Measurement of abnormal returns and its implication for the post-earnings announcement drift

This dissertation consists of three chapters and investigates the critical impact of selecting proper abnormal return measures for drawing inferences on one of the famous accounting anomalies – the post-earnings announcement drift (PEAD). In the process, other topics, such as earnings surprise metrics, measurement of stock return predictive power, and coefficient instability are also discussed. In the first chapter, I look at the PEAD from an event study perspective to see if the realization of unexpected earnings (event) actually changes the returns process. This approach leads to significantly different inferences for the PEAD from the usual buy-and-hold returns methodology. I demonstrate that there is a significant event-leading excess return exhibiting a systematic correlation with standardized unexpected earnings (SUE) measured as median analysts’ forecast error scaled by the stock price. The second chapter repeats the analysis using other earnings surprise measures. I find that the event-leading excess return, measured by the alpha of Carhart 4-factor model estimated in the pre-earnings announcement estimation period, is significant for all the different measures of earnings surprise (although sometimes the pattern looks relatively random especially when incorporating the price factors into the time series expected earnings model and using the standard deviation of the forecast errors as the deflator). In other words, there is a critical impact on the inferences regarding PEAD from using an event study perspective that is not specific to a particular earnings surprise measure. In the third chapter, I apply a three-period approach and stratify individual firms into deciles based on the predictive power of the Carhart 4-factor model, measured by the out-of-sample R-squared in an intermediate prediction period. Poor out-of-sample predictive power of an estimated model may indicate instability of the coefficients resulting in mismeasured abnormal returns and a spurious inference of anomalies such as the PEAD. I support this reasoning with empirical results that analysts’ forecast accuracy is associated with stock return predictive power, indicating it may also be linked to coefficient instability.Ph.D.Includes bibliographical reference