Calendar Anomalies (Sell in May, January Effect, Halloween)
Seasonal rules: sell in May, the Santa rally, the January small-cap effect, the Halloween indicator.
Most calendar effects shrank or vanished once published and arbitraged; the survivors are fragile to data-mining adjustments.
- Why it fails
- Many seasonal patterns are artifacts of data mining; once you correct for the number of rules tested, most lose significance, and those that were real tended to fade after publication.
- When / how it stopped
- Sullivan, Timmermann & White (2001) showed calendar effects largely fail to survive a data-mining correction; effects that did exist generally shrank or disappeared once published and traded.
Calendar anomalies are the folklore of markets: “sell in May and go away,” the Santa Claus rally, the January effect in small caps, the Halloween indicator. Each is a simple, memorable rule that says returns are predictably higher or lower in a particular slice of the year.
The most important critique is statistical. Sullivan, Timmermann & White (2001), in “Dangers of Data Mining,” subjected calendar effects to a test that accounts for the full universe of rules a researcher could have tried. Their finding: once you correct for that data snooping, the apparent significance of calendar effects largely disappears. With enough months, weekdays, and holiday windows to slice, some will look profitable purely by chance.
There is a real exception worth naming. Bouman & Jacobsen (2002), in the American Economic Review, documented the Halloween indicator — stronger returns from November through April than May through October — across a wide set of countries, a result harder to dismiss as a single-market fluke.
But even genuine seasonal effects share a common fate: once a pattern is published and widely traded, participants front-run the predictable date, and the edge compresses. A pattern everyone knows about and positions around stops paying.
The takeaway is not that seasonality is meaningless, but that the survivors are few and fragile to honest statistical correction. The reader can judge which, if any, clear that bar.
Sources
- Sullivan, Timmermann & White (2001), "Dangers of Data Mining: The Case of Calendar Effects in Stock Returns", Journal of Econometrics
- Bouman & Jacobsen (2002), "The Halloween Indicator, Sell in May and Go Away", American Economic Review
Frequently asked
Do calendar anomalies like "sell in May" still work in 2026?
Most do not hold up. Sullivan, Timmermann & White (2001) tested calendar rules while accounting for the universe of rules a researcher could have tried, and found that the apparent significance of calendar effects largely vanishes once you correct for that data snooping. The Halloween / "sell in May" pattern documented by Bouman & Jacobsen (2002) was a notable exception with broad international support — but even genuine seasonal effects tend to shrink after they are widely published and traded.
Why do most seasonal trading rules fail out of sample?
Because many were never real to begin with. With enough calendar slices — months, days of the week, holidays, turn-of-month windows — some will look profitable by chance. Sullivan, Timmermann & White (2001) show that once you adjust for the sheer number of rules that could have been mined, the statistical significance mostly evaporates. And any pattern that was real tends to get arbitraged away once enough traders position around the same predictable date.
Not investment advice — your mileage may vary, but the burden of proof is on the person claiming an edge. This entry describes general research and published evidence (or its absence), not a recommendation. See the full disclaimer.