Evaluating the Robustness of MarketAnomaly Evidence

This study investigates two ways that sample selection can impact inferences about market efficiency: (1) unintentional, nonrandom exclusion of observations because of lack of available data for some firm-years (“passive deletion”); and (2) the effects of extreme return observations. The analysis proposes and illustrates a set of simple diagnostic tests designed to assess the sensitivity of estimated hedge portfolio returns to sample composition. These diagnostics are applied to the accrual anomaly and the forecast-to-price anomaly and the results indicate that the forecast-to-price anomaly is not robust to the effects of passive deletion. Moreover, extreme returns — as few as 100 firm-year observations — appear to generate the observed abnormal returns to both strategies.