Changes in bonus contracts in the post-Sarbanes–Oxley era

Note that the optimal contract sometimes allows earnings management. This is because the benefit from encouraging the manager to increase effort may be greater than the cost to eliminate earnings management.

If these factors encourage the executive to engage in no earnings management, there is no additional risk to the executive. However, because some earnings management may still be optimal to the firm (see Evans and Sridhar 1996), there is additional risk on the executive.

Other papers supporting a decrease in earnings management include Lobo and Zhou (2006) and Chhaochharia and Grinstein (2007).

We consider bonus compensation, and not also stock option compensation, as our proxy for incentive-based compensation for several reasons. First, bonus contracts frequently tie payments explicitly to earnings. There is not similar evidence on such a relation between options and earnings. Further, academic research has documented that stock option grants are a function of other factors unrelated to earnings such as tax and financial reporting costs, retention, and increasing the convexity of compensation contracts (see Core et al. 2003 for a discussion).

Crocker and Slemrod (forthcoming) develop a model similar to that of Evans and Sridhar (1996), reflecting the trade-off between effort incentives and truthful reporting. They acknowledge that a setting such as Sarbanes–Oxley may reduce incentives to manage earnings and allow contracts to provide greater effort incentives.

We begin our study in 1996 despite full compensation data being available from 1993. Starting in 1996 allows us to have a more balanced panel pre- and post-Sarbanes–Oxley.

To identify such executives, we first identify all executives with a title containing the strings “CFO,” “fin,” “trea,” “accou,” “acct,” “chief fi,” or “contr.” We then manually review the titles for potential misclassification of a non-financial executive as a financial executive as the title could contain the search string but not be related to a financial executive. We also perform a manual search of other executives, on a test basis, for misclassifications.

If bonus targets are established using prior year data (for example, Merchant and Van der Stede 2003, p. 335, suggest that managers may use historical data in establishing financial targets), then prior-year earnings may be highly correlated with the target included in the bonus plan. Thus, our examination of managing earnings towards prior-year levels may shed light on managing earnings towards bonus targets, despite our inability to observe those targets directly.

See, for example, Matsunaga and Park (2001) and Dechow et al. (1994).

It is possible that managers manage earnings downward when unmanaged earnings are above an upper cutoff (income smoothing) or below a lower cutoff (big bath) to maximize a future bonus. However, in the setting around Sarbanes–Oxley, we believe that firms are less concerned with managerial efforts to smooth income or take a big bath and more concerned about managerial efforts to manage earnings upward, for the following reasons. First, 90% of SEC enforcement actions between 7/31/1997 and 6/30/2002 (the period leading up to Sarbanes–Oxley) related to revenues and expenses resulting from income-increasing earnings management (SEC 2003b). Second, the results in Healy (1985) suggest that firms do not penalize managers for managing earnings downward if unmanaged earnings are below the lower cutoff or above the upper cutoff. Finally, in our data, it is evident that there are ranges of earnings changes for which firms are less concerned about downward earnings management. Our results in Sect. 5 are consistent with firms not imposing large penalties for managing earnings downward for big baths or for smoothing.

We estimate the two-kink model by dividing the region of positive earnings changes into deciles, estimating Eq. 3 nine times allowing the right (middle) region to contain the one largest to nine largest (nine smallest to one smallest) deciles of earnings changes, and choosing the model with the lowest sum of squared residuals. To assess the robustness of the model fit, we also estimate the two-kink model by dividing the region of positive earnings changes into groups based on earnings changes (rather than equal size portfolios of firms). To pick the upper cut-off point for large positive earnings changes, we put the scaled positive earnings changes into increments of 0.025. Then, we allow the region of positive earnings changes to comprise all possible increments of 0.025 from 0.025 to 0.475, with the middle region having all other positive earnings change observations. We perform this procedure in both the pre- and post-periods, separately. The model with the lowest sum of squared residuals is the model with the large earnings changes defined as earnings changes greater than or equal to 0.05 for the pre-period and 0.125 for the post-period. We re-estimate Eq. 3 using this alternative model and our conclusions that the one-kink-log model is the model with the best fit in each period are unchanged.

An argument could be made that OLS may be appropriate for our model if zero bonuses do not represent censoring. Our conclusions are unchanged if we estimate using OLS.

We also estimate our regression by year to observe the pattern in the coefficient on positive earnings changes over time. The yearly coefficients are 0.035, 0.045, 0.052, 0.056, 0.052, and 0.064 in 1996–2001and 0.081, 0.080, 0.088, and 0.113 in 2002–2005. This pattern suggests a clear shift in the coefficient on positive earnings changes occurring in 2002.

We compute the marginal effects recognizing that the variable of interest is an interaction of a dichotomous variable and a continuous variable.

We calculate discretionary accruals from a cross-sectional version of the modified-Jones model (Subramanyam 1996; Bartov et al. 2000; Dechow et al. 1995; Guay et al. 1996). This model has been shown to provide a reasonable measure of discretionary accruals by Dechow et al. (1995) and Guay et al. (1996), although imperfect. Following Kasznik (1999), we include cash flow from operations in the model. We estimate that model using all firms on Compustat with total assets greater than $1 million, for each 2-digit SIC code, for each year 1996 through 2005. Estimating total accruals by SIC code and year allows any accounting rule changes affecting accruals across all firms in an industry to be captured in non-discretionary accruals. In untabulated results, we find that positive discretionary accruals decrease in the post-period, consistent with a reduction in reporting discretion.

The marginal effect of this variable, computed with consideration to the issue raised in Ai and Norton (2003) is also positive and significant at p < 0.10.

Decreases in positive discretionary accruals could reflect changes in the growth opportunities of the firm and changes in growth opportunities may reflect differences in the informativeness of earnings. To ensure that our proxy for changes in earnings management is not picking up changes in growth opportunities, we redo the analysis replacing the indicator variable capturing decline in positive discretionary accruals with an indicator variable capturing decline in growth opportunities (the decile of the greatest increase in book-to-market ratio from the pre- to the post-period). We find no difference in the weight place on positive earnings changes for firms with the greatest decline in growth opportunities relative to other firms.

We include the following control variables that might explain a firm’s decision to issue equity: deviation of executives’ equity portfolios from predicted incentive levels, whether a firm is cash constrained, proximity to constraints on issuing equity, book-to-market ratio, and concurrent stock returns.

Prior research (Carter et al. 2007) documents a reduction in stock option grants as firms began accounting for stock options using the fair-value method as defined in SFAS 123(R) (Financial Accounting Standards Board 2004).

We assume that increased weight on the effort incentive resulting from reduced financial reporting flexibility is similar to the increased weight that results from reduced noise in the performance measure.

The marginal effect of this variable, computed with consideration to the issue raised in Ai and Norton (2003) is also positive and significant at p < 0.01.

These results are robust to using the ratio of salary to bonuses as the dependent variable.