Abstract This article presents an empirical exploration of a specific market-inefficiency explanation for the observed post-earnings-announcement drift In stock prices. The research question Is whether or not the observed relation between unexpected earnings in quarter t and stock-price changes in quarter t+1 represents a failure of the market to characterize the time-series properties of earnings correctly. The article contributes to the existing literature in two ways. First, it provides corroborating evidence of the failure of the market to characterize the properties of the process underlying earnings correctly. Second, and more importantly. It directly tests the conjecture that this failure explains the post-announcement drift. Corroborating evidence of the market's failure to characterize the properties of the process underlying earnings correctly is derived from a distributed lag model; a LOGIT model is employed to regress unexpected earnings on their four most recent past realizations. The results indicate that the probability of positive unexpected earnings in quarter t+1 is increasing in its (lagged) values for quarters t through t-2 and decreasing In its (lagged) value for quarter t-3 . The ability of the model to predict future earnings changes and stock returns outside of the estimation period was examined as well. The results show that the model robustly predicts both one-quarter-ahead earnings changes and future (abnormal) stock returns. Furthermore, future stock returns remain predictable even after current unexpected earnings are controlled. This last relation further corroborates the incremental explanatory power of the lagged unexpected earnings over the unexpected earnings of the current quarter with respect to the post-announcement drift in stock prices. Although this predictability of future earnings changes and stock returns is consistent with the results of prior research (see, e.g., Foster 1977; Griffin 1977; Foster et al. 1984, table 1; Bernard and Thomas 1990, tables 1 and 5), there is an important difference between previous and present methodology. The tests here involve predicting future earnings changes and stock returns by using data from a holdout period, rather than by documenting correlations in the sample. Thus, the tests used here increase the confidence that the results are not driven by modeling or sampling errors. Evidence from this study and prior research is consistent with the conjecture that the market systematically errs in predicting one-quarter-ahead (quarter t+1) earnings and stock prices. Such an error implies that a drift would be observed during quarter t + 1 , as the market uses pre disclosure information to update expectations for earnings in quarter t+1. To my knowledge, the extent to which this error explains post-announcement drift has not been directly tested. I formally test the extent to which this systematic error in predicting future earnings explains the drift by exploring the relation between the drift In stock prices observed in quarter t+1 and unexpected earnings in quarter t, while controlling for the implications of past earnings for future earnings. In particular, cumulative abnormal returns (CARs) for the period commencing three days after the earnings announcements of quarter t and ending one day following earnings announcements for quarter t+1 were computed for four portfolios that were constructed on the basis of unexpected earnings in quarter t. Once the implications of past earnings for future earnings are controlled, the positive relation between unexpected earnings in quarter t and the drift in stock prices observed in quarter t+1 no longer exists. This finding suggests that the observed relation between unexpected earnings in quarter t and stock price changes in quarter t +1 is fully explained by a systematic error in forecasting earnings.
Eli Bartov (Wed,) studied this question.