The Impact of Accounting Conservatism on the Compensation Relevance of UK Earnings

Abstract

It has long been established that because of accounting conservatism, the contemporaneous correlation between returns and earnings is lower (higher) for good (bad) news firm-years. Meanwhile, prior analytical agency work suggests that the compensation role of accounting earnings is potentially greater (for tasks such as noise filtering and incentive balancing) when the contemporaneous correlation between earnings and returns is lower. Hence, since accounting conservatism implies that earnings have a lower correlation with returns in good news firm-years, the present paper hypothesises that UK CEO cash compensation exhibits a stronger (weaker) sensitivity to accounting earnings in good (bad) news firm-years. The empirical findings offer substantial support for this hypothesis and are robust to alternative estimation methodologies. In addition, the results appear not to be attributable to the well-established impact of earnings persistence on the compensation–earnings association. Overall, the findings are consistent with the notion that UK compensation committees appear to take cognisance of the impact of accounting conservatism when awarding earnings-based compensation. In addition, the present work offers additional insights into the nature of the interaction between the contracting and valuation roles of accounting numbers.

Acknowledgements

I thank Christof Beuselinck, John Board, Jan Bouwens, Ed Cahill, Vlastimil Cerný, Mark Hutchinson, Sofie van der Meulen, Franck Moraux, Martin Walker and participants at the 2004 European Accounting Congress in Prague for helpful comments on earlier drafts. I am also grateful for the constructive criticism offered by seminar participants at Tilburg University and Korea University Business School. This paper has benefited substantially from the comments of the joint guest editors, James Ohlson and in particular, Laurence van Lent. In addition, the insightful observations of two anonymous reviewers have made a very significant contribution to the development of the study. All remaining errors are the sole responsibility of the author. Finally, I thank departmental members at IGR, University of Rennes, France and CZU, Prague, Czech Republic, for extending their research facilities while this paper was being completed during the course of my recent sabbatical.

Notes

1. Bushman and Indjejikian (1993a, p. 4) state that ‘… a major implication of our analysis is that differences in the information content of earnings imply different relative weights on earnings and share price in the compensation contract’. Related ideas are discussed in a number of studies (e.g. Bushman and Indjejikian, 1993b; Feltham and Xie, 1994; Indjejikian, 1999; Lambert, 2001; Boschen et al., 2003).

2. Lambert and Larcker's (1987) seminal empirical work predicts and finds that the compensation sensitivity of earnings (relative to stock returns) is a decreasing function of the correlation between earnings and returns. However, in contrast to the present study, Lambert and Larcker (1987) do not focus on the differential response of compensation to earnings across good and bad news firm-years. Other aspects of their work are discussed in Sloan (1993).

3. It should also be noted that for some contracting goals a positive correlation between earnings and returns may be beneficial. However, as discussed in this paper, for tasks such as noise filtering and incentive balancing prior theoretical work implies an expanded compensation role for earnings when the returns–earnings relation is lower.

4. Barclay et al. (2005, p. 21) argue that ‘the use of earnings in CEO compensation contracts raises an important question about the objective that should guide accounting standard setters in making decisions about disclosure standards … if we make earnings contemporaneous with price changes to increase the correlation between the two, then we lose an important measure of delivered performance’.

5. I am grateful to an anonymous referee for this observation.

6. It is worth noting that while the present work is based upon a related empirical framework to that employed by Leone et al. (2006), they make no prediction with respect to the differential sensitivity of compensation to earnings for good news. Leone et al. (2006, p. 172) argue that because many non-compensation-related factors (e.g. debt contract considerations) influence the chosen level of conservatism, ‘the existence of these other factors likely causes the firm-value maximising level of accounting conservatism to differ from that chosen solely for efficient compensation contracting’. In contrast, this study takes the view that accounting plays multiple roles in compensation settings (e.g. Lambert, 2001) and many of the contracting benefits associated with the use of accounting emerge from the fact that earnings and returns aggregate information in different ways (e.g. Paul, 1992). From this perspective, the ex post settling up issue (which is the focus of the Leone et al. (2006) study) represents only one aspect of the overall compensation–returns–earnings paradigm.

7. Leone et al. (2006) also provide a second reason for not making any prediction in relation to the differential sensitivity of compensation to the accounting variable for good news. They argue (p. 172) that ‘most bonus plans use accounting earnings in a piece-wise linear fashion, thereby creating a mechanical asymmetric relation between cash compensation and accounting earnings’. However, it is important to recall that compensation committees award earnings-based compensation in both explicit and implicit ways (e.g. Bushman and Smith, 2001). As Baber et al. (1998, p. 170) point out ‘recent studies suggest that compensation committees adjust earnings-based performance measures when doing so improves incentive arrangements’. In any case, the impact of bonus plan design on the empirical results is considered in more detail later.

8. Firms in the ‘Financials’ sector are excluded from the pool of available companies since these firms tend to operate under different accounting and regulatory regimes to other UK companies.

9. In total, 202 non-financial firms had the required compensation, accounting and stock price data over the sample period. Of these 202, 118 did not have a December year-end throughout the sample period leaving a final sample of 84 firms. A wide cross section of firms with data over a significant time period is required in order to employ panel data estimation techniques. Data availability considerations meant that a ‘cut-off’ date prior to 1983 would have severely limited the sample size.

10. In cases where EPS _{t − 1} is negative, the per cent change in earnings variable calculated as [EPS _t − EPS _{t − 1}]/EPS _{t − 1} is incorrect. In such circumstances, Hamman (2002) recommends a correction whereby the absolute (rather than the actual) value of the denominator is used in the formula and that approach is adopted in the present study.

11. ‘Raw’ returns are used in the present context since market or industry-adjusted returns are more usually employed in frameworks testing theories relating to relative performance evaluation (e.g. Antle and Smith, 1986; Janakiraman et al., 1992; Aggarwal and Samwick, 1999b).

12. Pope and Walker (1999, p. 68) also suggest that ‘it can be argued that market-adjusted returns may provide a more reliable indicator of good versus bad news’. In this study, the annual market return for the sample firms is calculated as the unweighted mean of all sample firms' returns for each year.

13. Untabulated results indicate that that economy-wide inflation metrics are not a significant determinant of compensation changes for the present sample.

14. The present study controls for the impact of outliers by deleting the top and bottom 1% of observations for each variable (e.g. Collins et al., 1994).

15. In other words, if accounting conservatism does not lead to a significant differential information content in contemporaneous earnings and returns across good and bad news firm-years for the present sample, then any differences in the compensation sensitivity of earnings across good and bad news could not be reasonably attributed to the impact of conservatism.

16. Statistical inference from the results of the annual regression estimates is achieved using the following Z-statistic which is also employed in prior work (e.g. Barth, 1994; Lipe et al., 1998): Z = mean of t-statistic/(standard deviation of t-statistic/√(n − 1)). The t-statistics from the annual regression estimates are calculated using Huber–White standard errors with the Rogers (1993) adjustment for firm-level clustering. The results are similar if bad and good news are defined in terms of raw as opposed to market-adjusted stock returns.

17. Tests using the standard White (1980) approach suggest that heteroscedasticity is a problem for the pooled OLS tests. To test for autocorrelation at the panel level, the approach outlined in Wooldridge (2002) as implemented by Drukker (2003) is employed. For all the pooled OLS analyses this test suggests that autocorrelation is not evident. The inferences with respect to Hypothesis 1 remain unchanged when Newey–West (1987) standard errors (which adjust for both heteroscedasticity and first-order autocorrelation) are employed. Other tests reveal that the OLS results are not impacted significantly by collinearity (variance inflation factors) or omitted variable (Ramsey test) difficulties. Finally, the core inferences in Table 5 are unchanged if the analysis is repeated controlling for period-level as opposed to firm-level clustering (Rogers, 1993; Petersen, 2005).

18. The impact of earnings persistence is considered in more depth later in the paper.

19. For example, Matejka et al. (2005) show that, when compensating executives, firms tend to use non-financial or individual performance measures as opposed to earnings-based measures when earnings are negative.

20. The results from each of the sensitivity tests outlined here are available from the author on request. For the reasons discussed earlier, the analysis for the remainder of the paper focuses only on the findings with respect to Hypothesis 1 for the overall and positive ΔEPS samples.

21. The results with respect to Hypothesis 1 are also robust to the use of random effects and feasible generalised least squares (FGLS) panel estimation methodologies. An interesting aspect of the sensitivity tests is that the results for the ex post settling up hypothesis (Leone et al., 2006) are rather mixed depending on the particular estimation methodology employed. Indeed, for the annual OLS estimates, the R*D coefficient has the opposite sign to that predicted by Leone et al. (2006). These findings highlight the need for additional comparative cross-country research (e.g. Conyon and Murphy, 2000) on the compensation–performance association.

22. The advantage of the grouping approach used here is that it mitigates the effect of potential measurement error in the firm-specific estimates of earnings persistence. However, to further explore the impact of persistence, the firm-specific ERCs are included as an additional variable in equation (6) as is the interaction variable ΔEPS*D*ERC. For these tests, the ΔEPS*D coefficient is negative and significant (t = −2.50***) while the ΔEPS*D*ERC coefficient is positive and statistically insignificant (t = 0.98). The latter result offers further support for the notion that persistence does not significantly impact the reduced compensation sensitivity of earnings in bad news firm-years.