Perverse market rewards for meeting or beating earnings expectations^*

Abstract

Approximately 47 (43) percent of the observations in our sample receive negative (positive) market rewards when they meet (miss) earnings expectations. We define these phenomena as perverse market rewards (PMR). We find that the likelihood of PMR is increased when (i) firms use earnings and/or expectations management; (ii) earnings growth is negative (positive) when earnings expectations are met (missed); and (iii) ownership by transient (dedicated) institutional investors is high when earnings expectations are met (missed). In addition, we find that, when earnings expectations are met (missed), PMR appears to be an indicator of bad (good) future stock performance. Our study demonstrates that gratuitous participation in the ‘numbers game’ does not always result in the desired market rewards.

Keywords:

• Earnings expectations
• earnings management
• earnings growth
• expectations management
• institutional investors

Acknowledgements

Comments from Regina Anctil, Ozer Asdemir, Jennifer Blaskovich, Brooke Beyer, Marcus Caylor, Michael Clement, Bryan Cloyd, Scott Johnson, Bowe Hanson, Benjamin Landsford, Brian Shapiro, Sarah Stein, Wayne Thomas, Andrew Yi, Wen Yu, and seminar participants at Texas Tech and Virginia Tech are gratefully acknowledged.

Notes

* Accepted by Yue Ma upon recommendation by Junbo Wang

1. Yahoo’s earnings were $247 million, or 16 cents per share, compared with $187 million, or 13 cents per share, one year earlier. The reported amount fell short of Wall Street analysts’ estimates of 17 cents per share based on a survey by Thomson Financial.

2. Johnson and Zhao’s (2011) study is similar to ours in the sense that they examine the determinants of PMR. However, they neither consider earnings and/or expectations management nor the effects of institutional investors on PMR. More importantly, they use short-window returns around earnings announcements, whereas we use long-window returns consistent with prior research documenting MBE.

3. Following Matsumoto (2002), when PPE for the 1st through 3rd fiscal quarters is missing, but the data for the 4th fiscal quarter are not missing, we compute the year-to-year change in PPE and add to each of the interim quarters a proportional amount of this change based on the proportion of annual depreciation incurred in that quarter.

4. The data on institutional investor types are graciously provided by Brian Bushee; see http://acct.wharton.upenn.edu/faculty/bushee/IIclass.html.

5. Because consensus forecast is an average value, it is possible for the difference to be less than one cent. For example, if there are 2 forecasts, one for 1 cent and the other for 0, then the average could be between 0 and 1 cent.

6. Stock-split unadjusted I/B/E/S data are used due to a potential misspecification problem found by Payne and Thomas (2003). In addition, if a stock split occurs between the most recently issued forecast consensus and the earnings announcement, the stock price is adjusted accordingly.

7. Because we want to classify observations with zero surprise into one decile, the number of observations in each decile between positive and negative surprises is not equal. Observations with negative surprises are assigned to deciles 1 through 3, and observations with positive surprises are assigned to deciles 5 through 10.

8. The Cuzick test for trends is an extension of the Wilcoxson rank-sum test. The Cuzick test is more powerful than the Wilcoxon rank-sum test because it can test between two or more groups.

9. Note that FRCAGE is deflated by 365.

10. Insignificant results may be caused by a high correlation between these two variables with POSEARN. We check for any changes in the results after removing POSEARN from the model. However, we do not find evidence.

11. Instead of revenue growth, we use an indicator variable of whether revenue forecasts are met (FORREV). Because most companies do not have revenue forecasts available in I/B/E/S, total observations are significantly reduced to 10,944 (a loss of 93% of total observations). In contrast to the insignificant result of revenue growth in MBE sample, the coefficient on FORREV is negatively significant. However, POSDA and %DED become insignificant potentially due to the loss of statistical power. The results of non-MBE sample are qualitatively unchanged when we use FORREV.

12. We conduct several sensitivity tests. First, we use the absolute values of forecasted EPS as the deflator for |SURPRISE| instead of beginning price, and find that our results are robust. Second, we winsorize continuous variables instead of trimming the top and bottom 1% of the sample, and find that this does not change the tenor of our results. In addition, the results remain largely the same without trimming.