We explore the causes of stock return anomalies by studying the effects of new information. Using a sample of 97 anomalies from published academic studies, we find that anomaly returns are 500% higher on earnings announcement days, and 200% higher on corporate news days. The size of these differences is difficult to explain with time-varying risk exposures. Anomalies are unlikely to be explained by data mining; pseudo-anomaly portfolios, which are data-mined portfolios with the same monthly returns as genuine anomaly portfolios, are less affected by new information. We further find that anomaly variables predict analyst forecast errors. Taken in their entirety, our findings point to errors in expectations as the source of anomaly returns. Joint with Joseph Engelberg and David McLean.