Calendar anomalies: Real patterns or data-mining artifacts?

This paper reexamines well-known calendar anomalies in equity returns with the objective of determining whether these effects are genuine return regularities or artifacts of data mining. We study four canonical anomaly families—the day-of-the-week, week-of-the month, month-of-the-year (January), and Sell-in-May effects—using U.S. equity data, and we also evaluate international evidence for the Sell-in-May effect. Our methodology explicitly accounts for the selection inherent in the discovery of calendar anomalies by applying data-mining-adjusted bootstrap tests that mimic a researcher’s within-family search across alternative calendar definitions. Our results show that, after correcting for data mining, the day-of-the-week, week-of-the-month, and month-of-the-year anomalies remain statistically significant in the full sample, with the strongest evidence coming from the earlier part of the sample, indicating that these effects were not merely statistical artifacts. At the same time, we find that these anomalies largely disappear in later subsamples beginning in the early 1990s. In contrast, the Sell-in-May effect proves more resilient: international evidence remains robust to data-mining concerns. Overall, the findings suggest that several calendar anomalies were real features of historical return data, even though their economic relevance has diminished in more recent decades.