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

Abstract

If overstatements were a symptom of the agency conflict, pay-for-performance sensitivities should have increased in response to the additional penalties for misreporting imposed by SOX. Our finding of their decrease is inconsistent with the view that overstatements were an unintended consequence of incentive pay prior to 2002. To corroborate our interpretation, we show that (i) CEO pay-for-performance sensitivities are higher among firms whose shareholders stand to benefit from overstatements; (ii) this cross-sectional relationship weakens significantly after SOX; and (iii) the within-firm decrease in pay-for-performance sensitivity is most pronounced among firms with high pre-SOX shareholder benefits from overstatements.

Appendices

Appendix 1: Proofs

Proof of Proposition 1

From (5),

$$\begin{aligned} \frac{\partial \beta ^{*}}{\partial \rho }= & \frac{2\eta ^{2}\left(1-\lambda \eta (1-\theta )\left (1+r(\sigma _{a}^{2}+\sigma _{m}^{2})\right)\right)}{\left (1+2\eta ^{2}\rho \left(1+r(\sigma _{a}^{2}+\sigma _{m}^{2})\right)\right)^{2}}\ge 0 \\ \Longleftrightarrow & \lambda \le \frac{1}{\eta (1-\theta )(1+r(\sigma _{a}^{2}+\sigma _{m}^{2}))}. \end{aligned}$$

(10)

Since λ ≤ 1, $\frac{\partial \beta ^{*}}{\partial \rho }$ is always non-negative if $\frac{1}{(1-\theta )(1+r(\sigma _{a}^{2}+\sigma _{m}^{2}))}\ge 1,$ that is, if $\theta \ge \frac{\eta r(\sigma _{a}^{2}+\sigma _{m}^{2})-1}{\eta r(\sigma _{a}^{2}+\sigma _{m}^{2})}$.

If $\theta < \frac{\eta r(\sigma _{a}^{2}+\sigma _{m}^{2})-1}{\eta r(\sigma _{a}^{2}+\sigma _{m}^{2})}$, however, $\frac{\partial \beta ^{*}}{\partial \rho }<0$ if and only if $\lambda >\frac{1}{\eta (1-\theta )(1+r(\sigma _{a}^{2}+\sigma _{m}^{2}))}$.

Proof of Proposition 2

• (i) From (5), it is straightforward to show that

  $$\frac{\partial \beta ^{*}}{\partial \lambda }=\frac{\frac{1-\theta }{2\eta \rho }}{1+\frac{1}{2 \eta ^{2} \rho }+r(\sigma _{a}^{2}+\sigma _{m}^{2})}>0.$$

  (11)

• (ii) From (10),

  $$\frac{\partial ^{2}\beta ^{*}}{\partial \lambda \partial \rho } =-\frac{2\eta ^{3}(1-\theta )\big (1+r(\sigma _{a}^{2}+\sigma _{m}^{2})\big )}{\Big (1+2\eta ^{2}\rho \big (1+r(\sigma _{a}^{2}+\sigma _{m}^{2})\big )\Big )^{2}}<0.$$

  (12)

Appendix 2: Details on Calculating PPS

We construct the incentive measure following Core and Guay (2002). In particular, we compute the dollar change in executives’ stock and option holdings for a hypothetical one percent change in firm value [we call this variable pay-for-performance sensitivity (PPS)]. We separately calculate PPS for newly granted options, previously granted exercisable and unexercisable options, and stock holdings. Measuring PPS requires six inputs: the risk-free rate, stock price volatility, dividend yield, time-to-maturity, stock price, and number of options granted or held. All variables except for the risk-free rate can be obtained from Execucomp, either directly (e.g., dividend yield and volatility, stock price) or indirectly (time-to-maturity, number of options held).

Following the Execucomp convention in calculating option grant values, we winsorize volatility and dividend yields within each fiscal year. The largest and smallest values are least likely to be good representations of expectations about their future values. We replace missing values of the 3-year average dividend yield (bs_yield) with current dividend yields, missing values for volatility (bs_volat) with the Execucomp sample mean, and missing values for exercise price (expric) with either the market price (mktpric) or the average of the fiscal-year-end closing price (prccf) and the closing price discounted by total shareholder returns that year (trs1yr). We also observe that firms who make only one grant to an executive within a fiscal year often only report the total number of options granted (soptgrnt), but not the number of options in that grant (numsecur). We estimate maturity to be the difference between exercise date and grant date. Missing values are assumed to be 10 years. Some maturities are computed to be 0 years, so we replace those with 1 year. We also value the options at the end of the fiscal year, not at the time of the grant to make all values comparable and current at fiscal year end. Finally, we weight the individual grants’ deltas by the grant values to each executive within each year to compute PPS from new option grants for each executive-firm-year.

Estimating the inputs for previous grants is harder. Information on the characteristics of past option grants is not available. For example, the number and value of unexercisable options are available, but we do not know the composition of the unexercisable options from previous grants. Similarly, for exercisable options, we do not know which previously granted options were exercised by the executives and which ones were kept in the portfolio. However, Core and Guay’s main contribution lies in showing that imputing the missing characteristics yields a very close approximation to hand-collected, full-information option portfolios. Unfortunately, the documentation in Core and Guay does not allow us to replicate their imputation strategy directly. We encounter a number of problems. For example, the reported value of (un)exercisable options pertains only to in-the-money options, but the number of (un)exercisable options also includes out-of-the-money options. Furthermore, adjusting the value and number of unexercisable options for current year option grants imply that about half of our observations would end up with negative values. We assume that the reported number of unexercisable options held includes newly granted options, unless the number of options granted exceeds the holdings. Similar to our approach for newly granted options, we estimate the exercise price for previously granted options by appropriately discounting the adjusted fiscal-year end stock price by total shareholder returns (trs3yr). The maturity of unexercisable options is assumed to be one year less than the maturity of any option grant in the previous year, or 9 years if no options were granted in the previous year. The maturity of exercisable options is assumed to be 3 years less than that of unexercisable options.

Appendix 3: Robustness Checks

Representativeness of the Mean Effect

In Tables 3 and 4 we report results from firm-fixed-effects regressions that estimate the mean change in CEO pay-for-performance sensitivities from before to after SOX. To ensure that our results are representative of the typical firm in the sample instead of being driven by large changes in a few firms, we also estimate median regressions. The results are presented in Table 8. We purge firm-fixed effects by demeaning all variables.³² The estimated median change in PPS from before to after SOX is almost identical to the mean effect. We conclude that the change in PPS is pervasive and representative of the typical firm in our sample.

Table 8 The change in PPS around SOX: median regression

Bonus Pay

Our measures of the performance sensitivity of CEO compensation emphasize CEOs’ wealth gains from stock and option holdings. In practice, however, other forms of pay, such as bonuses, are also tied to firm performance and can thus provide incentives for overstatements. Our first measure of the level of performance sensitivity—log(PPS)—completely ignores CEOs’ bonus compensation. Although our second measure—log(PPS-ratio)—includes bonuses, it assumes that bonuses provide CEOs with fewer incentives to overstate performance than stock and option holdings. To rule out the possibility that CEO incentive pay shifted from PPS to bonus pay around SOX without affecting the link between total CEO pay and firm performance, we take an alternative approach offered in the prior literature on CEO pay to estimate how the performance sensitivity of CEO pay has changed around SOX. We regress bonus pay and total CEO pay on two measures of firm performance: return on assets and firm stock returns. We also interact the performance measures with the post-SOX dummy to allow for changes in the performance sensitivity of CEO pay:

$$\begin{aligned} \textit{pay}_{it}= & \tau _{1}\ \textit{performance}_{it} + \tau _{2}\ D(t\ge 0)_{t} \times \textit{performance}_{it} + \tau _{3}\ D(t\ge 0)_{t} \\ \quad\quad & + \alpha _{0}+\sum _{j=1}^{k}\alpha _{j}X_{jit}+\upsilon _i+\epsilon _{it},\end{aligned}$$

(13)

where $D(t\ge 0)_{t}$ is a dummy set to one for fiscal years 2002–2005. The interaction term captures whether the link between pay and performance has strengthened or weakened from before to after SOX. Again, we estimate heteroskedasticity-robust standard errors, clustered at the firm-period level.

The results are displayed in Table 9. In column 1, we use bonus pay as the dependent variable. In column 3, we use total CEO pay as the dependent variable, which includes the flow of compensation (such as salary, bonus, stock and option grants), as well as changes in the value of CEOs’ stock and option holdings. Column 2 is the in-between case, where we exclude bonuses from total pay. We use the dollar value of bonus and total pay (in ${\$}$ mill.) instead of their logarithmic values, because the dollar amounts are zero or negative in a non-negligible fraction of observations. To alleviate the concern that outliers severely affect the magnitude of our estimates, we winsorize the pay and performance measures at the top and bottom percentile.

Table 9 The changing link between CEO pay and firm performance

The result for bonus pay confirms that incentive pay has in fact shifted from stocks and options toward bonus pay. We estimate that bonus pay has increased by ${\$}$166,000 around SOX on average. Furthermore, bonus pay does increase with return on assets (accounting performance) and with firm stock returns (market performance). Most interesting, however, is the finding that the accounting-performance sensitivity decreases around SOX, while the market-performance sensitivity of bonus pay increases. This shift towards bonus pay and its increasing market-performance sensitivity suggest that our earlier results based on log(PPS) overstate the true decrease in the performance sensitivity of CEO pay.

Turning to total pay in column 3, we find that it primarily responds to firms’ market performance. The economic magnitude of its performance sensitivity swamps the wealth effects from bonus pay.³³ More importantly, the performance sensitivity of total pay decreases sharply around SOX by almost half.

How large is the contribution of bonus pay on the performance sensitivity of total pay? Comparing the coefficient estimates between columns 2 and 3 reveals that bonus pay has a negligible effect; most of the performance sensitivity derives from the firm’s stock return, and the coefficient estimate is almost unchanged when bonus is excluded from total pay (and bonus has a minor impact on the sensitivity to accounting performance). We conclude that the declining performance sensitivity of stock and option holdings outweighs the increasing weight placed on bonus pay and its increasing market-performance sensitivity.

Contemporaneous Corporate Governance Reforms

In this section, we assess the robustness of our main findings to excluding firms affected by the contemporaneous board independence requirements.

We use board data provided by Riskmetrics to determine firms’ compliance status. We match the Riskmetrics observation to the fiscal year into which the board meeting date falls. We classify boards as compliant or non-compliant based on their board independence in fiscal year 2002, the year prior to the rule change. Following Chhaochharia and Grinstein (2009), we reclassify directors as independent when their employment relationship terminated three or more years ago to reconcile the differences in how Riskmetrics and the NYSE/Nasdaq listing standards define independence. Of our 857 sample firms, we classify 138 as non-compliant, and lack board data for 77.

The new listing requirements had a noticeable impact on board independence. The change in board independence is evident in Table 2, panel A. Firms that were failing the new director independence standards in the year prior to those rules going into effect, improved their governance drastically over the following years. In the non-compliant firms, only 42 % of directors were independent before the new rules, but independence increased by 10 percentage points within one year and by 20 percentage points by 2005. On the other hand, firms that already met the requirements show an increase of only 3 percentage points from 2002 to 2005. The fraction of compliant boards in our sample jumps from 82 % in 2002 to 93 % in 2004.

We allow the effect of SOX on PPS to differ between compliant and non-compliant firms by estimating regression (6) separately for compliers and non-compliers. In addition, we add various measures of board characteristics that might either affect PPS (e.g., board ownership, tenure and age of directors) or vary systematically around SOX (e.g., board size, board independence, and the number of directorships of board members) as control variables. Table 10 displays the results. PPS decreases in compliant firms, which suggests that even independent boards emphasize market values over fundamental values. Thus, we should not expect independent boards to be effective monitors of overstatements.

Table 10 The change in incentives around SOX: controlling for changes in board characteristics

The economic magnitude of the change in log(PPS) is three times larger for non-compliers than compliers, but not for incentives measured as log(PPS-ratio). The difference between the SOX effects, however, is not statistically significant for log(PPS) with a p-value of 29.9 %.

Note that the estimate of the decrease in PPS around SOX for compliant firms isolates the effect of shareholder myopia from changes in board independence. In contrast, the estimate for non-compliant firms captures both myopia and changes in board independence. Under the assumption that the effect of myopia is the same for compliant and non-compliant firms, our estimates suggest that board independence leads to an economically significant decrease in the performance sensitivity of CEO pay [at least if measured as log(PPS)]. There are at least two possible explanations for this finding. First, the decrease in PPS in non-compliant firms is consistent with the view that oversight and incentive pay are substitutes (Holmström 1979). The large decrease could thus reflect not just the change in the cost of overstatement, but also the improvement in the quality of corporate governance. Second, as suggested by Bertrand and Mullainathan (2001), non-independent boards may not have been setting or enforcing optimal incentive contracts. Therefore, the large decrease in PPS could also be attributable to a regime shift from managerial skimming to optimal contracting. That compliant and non-compliant boards differ in the allocation of power between managers and shareholder becomes evident when one compares board ownership. The mean pre-SOX ownership of compliant boards is only 6.4 %, but 20.0 % in non-compliant firms. Ownership by independent directors, however, is much smaller in magnitude and about equal at 1.1 % in compliant firms and 1.2 % in non-compliant firms.

Our model does not contain a parameter for board independence and thus does not offer predictions about the effect of independence (and its interaction with SBO) on CEO incentive pay. In light of the alternative views on the role of board independence, we simply replicate the empirical tests of Hypotheses (2) and (3) for the subsample of firms in compliance with the new board independence requirement in fiscal year 2002 and control for the various board characteristics mentioned previously. The results remain qualitatively, and in most cases even quantitatively, unchanged, as shown in Tables 11 and 12. We conclude that our findings are not attributable to the contemporaneous changes in board characteristics.

Table 11 The link between CEO pay-for-performance and shareholder benefits from overstatements in the cross-section: controlling for changes in board characteristics

Table 12 The impact of shareholder benefits from overstatements on the change in incentives around SOX: controlling for changes in board characteristics

Table 13 The change in incentives around SOX: non-high-tech vs. high-tech firms

Table 14 The link between CEO pay-for-performance and shareholder benefits from overstatements in the cross-section: excluding high-tech firms

Table 15 The impact of shareholder benefits from overstatements on the change in incentives around SOX: excluding high-tech firms