What Does CEOs’ Pay-for-Performance Reveal About Shareholders’ Attitude Toward Earnings Overstatements?

Firm values are based on managers’ reported earnings. A firm’s market value can temporarily exceed its fundamental value, because managers can report inflated earnings in excess of fundamental, or true, earnings.

SOX increased the cost to CEOs for overstating earnings by (i) increasing the limits on financial penalties and prison terms for financial misrepresentation; (ii) requiring CEOs to reimburse any incentive-based compensation or profit from the sale of stock received within 12 months after the misreporting if there is an accounting restatement as a result of misconduct; (iii) providing an additional 776 million in funding to the Securities and Exchange Commission (SEC) to step up its monitoring and enforcement efforts; and numerous other provisions.

Our conjecture that a change in the cost of overstatements affects the optimal level of PPS is corroborated by Karpoff et al. (2008a, b). The authors show that managers and firms suffer substantial penalties for financial misrepresentation if caught, which suggests that those consequences factor into managers’ and shareholders’ choices.

Our findings are qualitatively robust to numerous robustness checks, including those pertaining to variable definitions and measurement (e.g., pre/post-SOX period, treatment of bonus pay, flow vs. level of incentive pay) and to various sample restrictions (e.g., non-high-tech firms, firms that were compliant with contemporaneous governance regulations, market-value-matched pre/post-SOX firm-year-pairs).

Two notable exceptions are Erickson et al. (2006) and Armstrong et al. (2010). However, Erickson et al. (2006) base their study on a small number of accounting frauds that likely reflect idiosyncratic managerial expropriation; and the sample period of Armstrong et al. (2010) spans pre- and post-SOX years and their results are not robust when restricted to the pre-SOX period.

See “Discussion” section for further discussion of alternative explanations of our findings.

Their result is likely driven by their research design. They estimate the SOX effect after controlling for year and industry effects, but not for firm-fixed effects.

For example, see Schipper (1989), Beneish (2001), Dechow and Dichev (2002), Kothari et al. (2005) and Ball and Shivakumar (2006) on the accruals debate, Durtschi and Easton (2005) on forecast errors, and Dechow et al. (1996), Burns and Kedia (2006), Hennes et al. (2008), Peng and Röell (2008), Wang (2012) on enforcement actions, restatements, and litigation.

Throughout the paper, we ignore the possible agency problem between the shareholders and the board. Allowing such agency problem in this model would be an interesting topic for future research.

For example, D’Avolio (2002), Geczy et al. (2002) and Jones and Lamont (2002) provide empirical evidence that it is costly to short sell stocks.

This assumption reflects the usual stock- and option-based compensation packages for CEOs; the model’s predictions are unchanged if the agent’s compensation were tied to an accounting performance measure instead. The crucial assumption in our model is that the agent’s report on fundamental performance is not verifiable. For example, the agent may only know the probability distribution of the true performance, and can only report the mean of the distribution. Then, the agent is unlikely to become liable for the report. This assumption allows us to circumvent the revelation mechanism, as discussed in Dye (1988) and Crocker and Slemrod (2005).

For recent attempts to characterize general non-linear contracts, see Hemmer et al. (2000) and Crocker and Slemrod (2005).

Kwon and Yeo (2009) show that there is another, more complex equilibrium where market expectation is a strictly increasing function of reported performance. Such an equilibrium becomes quickly untractable in this paper, but the qualitative results of this paper should hold in that equilibrium too.

We focus on institutional investors rather than retail shareholders for two reasons: (i) they have greater influence on firm policies and characteristics due to larger ownership stakes and greater financial sophistication, and (ii) they dominate the shareholder base of our sample firms.

The classification is available at http://​acct.​wharton.​upenn.​edu/​faculty/​bushee/​IIclass.​html.

We add fiscal year 1999 to balance the number of pre- and post-SOX years, with 2002 being the transition year.

We tested two definitions of high-tech: (i) firms in the communications, computer, electrical, and electronic equipment industries based on the Fama-French 48-industry classification and (ii) firms with SIC codes 3570–3572, 3576–3577, 3661, 3674, 4812–4813, 5045, 5961, 7370–7373 as in Ferri et al. (2006). See estimates in Appendix Tables 13, 14, 15.

The results are also qualitatively robust to excluding firms in the tails of the distribution of the change in average PPS from the pre- to post-SOX period.

SOX was passed in July 2002 in response to the large corporate scandals in the preceding year (e.g., Enron, Tyco, Worldcom). We assume that fiscal year 2003 falls into the post-SOX period, as its begin date falls between June 2002 and May 2003. To the extent that the expected cost of overstatements increased prior to the adoption of SOX (e.g., through anticipated regulatory changes or higher scrutiny by investors and enforcement agencies), effects on incentive pay can already be visible in earlier years.

Controlling for R&D and cash constraints as predictors of option usage does not materially affect our estimates.

Optimally designed pay-for-performance packages encompass a variety of performance measures (e.g., see Merchant 2006 on the benefits and drawbacks of various market and accounting performance measures and Schiehll and Bellavance 2009 on non-financial performance measures). However, we focus on pay-for-performance arising from stock and option holdings for the following reasons: (i) lack of granular data on incentive plans and performance measures, (ii) their economic significance in CEO compensation packages among our sample firms during the sample period, and (iii) empirical evidence linking specifically stock and option holdings to earnings overstatements that culminated in the passage of SOX.

We calculate the percentage change as \({\rm exp}(-0.232)-1=20.7\%\). We calculate the dollar change by multiplying the percentage change with the mean and median values of PPS of the sample firms before SOX.

Clustering only at the firm level does not materially alter any estimated standard errors.

In contrast to the strong results in the cross section, we uncover no systematic relationship between within-firm variation in the SBO-scores and PPS. The within-firm variation comes from only 3 years in the pre-SOX period, and 4 years in the post-SOX period, but not from across the periods. Given the limited number of observations per firm over time and the lower within-variation in SBO-scores mentioned previously, this finding is not surprising.

Dikolli et al. (2009) also find that bonuses—which capture only a fraction of total incentive pay—are more sensitive to stock returns than earnings, and equity grants are larger when transient institutional ownership is high. The authors interpret these findings as evidence that CEO incentive contracts are designed to offset myopia.

The results remain qualitatively unchanged if we use continuous pre-SOX averages of the proxies for shareholder benefits instead of their dummy versions, or consider only the top quartile of each SBO-score as benefitting shareholders.

We believe that our implicit assumption that managers’ and shareholders’ cost of overstatement increased proportionately is rather conservative. The main cost of overstatements to shareholders stems from loss of reputation rather than legal and regulatory penalties if caught (Karpoff et al. 2008b). Since SOX does not alter the market’s perception of reputation loss, its main direct effect on shareholders’ expected cost comes from a greater enforcement effort (i.e., the probability of getting caught). Managers are equally affected by the increase in the probability of getting caught, but are subject to the numerous additional new or increased personal penalties that SOX imposes on them.

Including firms with less accurately matched observations does not affect our findings.

One might argue that shareholder proposals to vote on option expensing forced firms to expense options, and consequently affected pay-for-performance. However, our results are robust to excluding all firms with such shareholder proposals during the 2003 and 2004 proxy seasons (as identified in Table 1 in Ferri et al. (2006).

We do not speak to the efficiency of overstatements. On the one hand, earnings overstatements can distort investment decisions. If firms appear more profitable than they are, managers invest in insufficiently profitable projects to mimic investment and employment of truly profitable firms (as documented in Kedia and Philippon (2009), for example). On the other hand, Shleifer and Vishny (1990) argue that short-term arbitrage being cheaper than long-term arbitrage leads to firms focusing on short-term assets to avoid prolonged underpricing. That is, firms may avoid long-term investments with positive net present values, because of fear of underpricing. Therefore, contracts that encourage CEOs to avoid underpricing by inflating earnings could in fact alleviate underinvestment in long-term assets.

First-differencing instead of demeaning does not materially affect the results.

As CEO pay is highly skewed, estimated mean effects are not representative of the typical firm. Using median regressions reduces the magnitude of the estimates by factors ranging from 2 to 4, but the qualitative findings do not change.