Calendar Anomalies: Are January, Turn-of-Month and Sell in May Real?

Calendar anomalies are return patterns tied to the calendar: the month of the year, the day of the month, the day of the week, holidays, or the half of the year. The honest summary of the research is mixed. Some patterns, especially the “Sell in May” (Halloween) effect, look persistent in large international samples, while data-snooping tests and a 300-year UK study argue that many others are fragile and sample-specific. This article reviews the main anomalies, what each key study actually found, and why the debate is still open.

Calendar Anomalies by the Numbers

What the main calendar anomalies claim

Each calendar anomaly says that average returns differ systematically depending on when you hold. The main ones are:

• The January effect (seasonality around the turn of the year).
• The turn-of-the-month effect (returns concentrated around month boundaries).
• The weekend effect (returns differing around the turn of the week).
• Holiday effects (returns differing around market holidays).
• The Halloween indicator, popularly “Sell in May and go away” (winter months, November to April, versus summer months, May to October).

None of these has an obvious reason to exist. Companies do not earn more cash because the calendar page turned, so a durable pattern would need an explanation from flows, behavior or institutions. The related article on behavioral finance covers some of those candidate explanations; the sections below focus on whether the patterns are real in the first place.

The early papers: January and the weekend

The seasonality literature in our sources begins in 1976. Rozeff and Kinney (1976) published an early paper on seasonality in stock returns in the Journal of Financial Economics, titled “Capital market seasonality: The case of stock returns”. Keim (1983) followed with “Size-related anomalies and stock return seasonality”, connecting seasonality to firm size. French (1980), “Stock returns and the weekend effect”, is the standard citation for the weekend effect.

We cite these papers by title and year only. We could not retrieve their abstracts or full text when preparing this article, so we do not report their numerical findings. If you want the magnitudes, read the original papers rather than relying on secondary summaries, which vary.

Turn of the month and 90 years of Dow data

Ariel (1987) examined the pattern within the month. He reported that mean stock returns are positive only for days immediately before and during the first half of calendar months, and indistinguishable from zero for days during the last half. He said the effect is independent of the January effect, and that it appears to be caused by a shift in the mean of the return distribution between the first and second halves of the month.

Lakonishok and Smidt (1988) asked the title question directly: “Are Seasonal Anomalies Real? A Ninety-Year Perspective”. They used 90 years of daily data on the Dow Jones Industrial Average to test for persistent seasonal patterns. They found evidence of persistently anomalous returns around the turn of the week, around the turn of the month, around the turn of the year, and around holidays. That is the strongest case in favor of the older anomalies, and it covers four of the five patterns listed above in a single long sample. They also considered methodological issues in testing seasonality, which is where later critics pushed.

We did not verify the size of these effects from the full texts, so we do not quote magnitudes for either paper.

The Halloween indicator: the strongest case for persistence

Bouman and Jacobsen (2002) published “The Halloween Indicator, ‘Sell in May and Go Away’: Another Puzzle” in the American Economic Review. Their abstract was truncated in every source we could reach, so we rely on the description in Jacobsen and Zhang: Bouman and Jacobsen found that returns during winter (November through April) are significantly higher than during summer (May through October) in 36 of the 37 countries in their study. That count is Jacobsen and Zhang’s account, not a direct quotation of the AER paper.

Jacobsen and Zhang then widened the net. Their paper is a working paper (SSRN 2154873; the title varies across versions, and we could not confirm a peer-reviewed publication). They extended the test to all 109 stock markets with price indices, with UK data starting in 1693. Winter returns exceeded summer returns by 4.5% (t = 11.42) over 56,679 monthly observations spanning 319 years. Returns for November to April were higher than for May to October in 82 of 109 countries, with the difference statistically significant in 35 countries (and in 2 for the reverse). A later passage of the same paper gives 81 of 108, so the counts vary slightly within the document.

They also ran an out-of-sample test on the 37 original countries from September 1998 to November 2011. All 37 showed positive point estimates, and 15 were statistically significant, which led them to conclude the effect “seems not to have weakened”. Pooling all country data, they report significantly negative summer excess returns (−1.2% based on 33,348 monthly returns), and they frame the result as a challenge to the positive risk-return trade-off. Their design, covering the full history of all available markets, is in their words “probably the best safeguard against data mining and sample selection bias”.

The effect is controversial. Because the evidence comes from an SSRN working paper, and the 36-of-37 count is second-hand, treat it as a serious claim that has not been through the same checks as a published result.

The skeptics: data snooping and three centuries of UK data

Sullivan, Timmermann and White (2001) tested calendar rules using 100 years of daily data and a new bootstrap designed to measure data-snooping distortions. Their conclusion, in the working-paper abstract, is that “although nominal P-values of individual calendar rules are extremely significant, once evaluated in the context of the full universe from which such rules were drawn, calendar effects no longer remain significant”. They describe the presence of calendar effects as “a striking example of a data-driven discovery” that follows from intensive prior search. The quoted wording comes from the 1998 working-paper version; the journal abstract may differ slightly, and their sample runs to the mid-to-late 1990s.

The logic is simple. If researchers test hundreds of calendar rules, some will look significant by chance, and the ones that get published are the ones that did. A single rule’s p-value ignores how many rules were tried. This is the same trap discussed in overfitting in backtests.

Zhang and Jacobsen (2013) came at the problem from the other side: more history. Using over 300 years of UK stock returns, they found that well-known monthly seasonals are sample-specific. The January effect, for example, only emerges around 1830. Their overall conclusion is that monthly seasonals “might simply be in the eye of the beholder”, and that most months have had “50 years of fame”. Interestingly, the same author, Jacobsen, is on both sides of the ledger: his Halloween work claims persistence while this paper warns about monthly seasonals in general.

Do anomalies decay after publication?

A separate literature looks at what happens to predictors after they become known. McLean and Pontiff (2016) studied 97 variables shown to predict cross-sectional stock returns and found portfolio returns are 26% lower out-of-sample and 58% lower post-publication; the out-of-sample decline is an upper-bound estimate of data-mining effects. That is evidence about return predictors in general, mostly not calendar effects, so it must not be read as a measurement of calendar anomalies specifically. We did not find a fetched source measuring post-publication decay for the turn-of-month, weekend or January effects, so we make no claim either way.

What a practitioner can take from this

Treat calendar rules as hypotheses, not edges. Four practical points follow from the evidence above.

• Count your trials. If you pick one rule after scanning many, adjust for the search. Sullivan, Timmermann and White show how large that adjustment can be.
• Prefer long, wide samples. Lakonishok and Smidt used 90 years of data; Jacobsen and Zhang used 109 markets. A pattern that exists only in one index and one era deserves suspicion.
• Test out of sample. The Halloween result is more credible because it was rechecked on 1998–2011 data. Methods such as walk-forward optimization apply the same discipline to your own rules.
• Account for costs and flows. A seasonal rule that trades in and out several times a year pays costs each time, and predictable calendar buying such as pension fund rebalancing flows may be crowded. Our article on backtesting trading strategies covers how to test net of those frictions.

Seasonality also shows up in other data. Our look at insider trading seasonality asks when insiders buy most, which is a different question from when the market earns most.

Limitations

The evidence base here is thinner than the topic’s popularity suggests. The findings of Rozeff and Kinney (1976), Keim (1983) and French (1980) were not verified from primary text, so this article reports none of their numbers. The magnitudes of Lakonishok and Smidt’s results were not verified either. The Bouman and Jacobsen count of 36 of 37 countries comes through Jacobsen and Zhang, not from the AER abstract, and the Jacobsen and Zhang paper is a working paper with slightly inconsistent counts across sections (82 of 109 versus 81 of 108).

Statistical significance is not the same as tradable profit. None of the figures quoted here is a transaction-cost-adjusted strategy return, so nothing above shows that a calendar rule makes money after costs and taxes. The Sullivan, Timmermann and White tests end in the 1990s and cover Dow and S&P calendar rules, so they do not speak to what has happened since. Nothing in this article is investment advice.

FAQ

What are calendar anomalies in the stock market?

They are patterns in average returns linked to the calendar, such as the January effect, the turn-of-the-month effect, the weekend effect, holiday effects and the Halloween (“Sell in May”) pattern. Lakonishok and Smidt (1988) found persistently anomalous returns around the turn of the week, month and year, and around holidays, in 90 years of Dow data.

Does “Sell in May and go away” work?

According to Jacobsen and Zhang’s working paper, November to April beat May to October in 82 of 109 markets, and the effect had not weakened in an out-of-sample test from 1998 to 2011. It is the most persistent-looking calendar anomaly in our sources, but it comes from a working paper, is disputed, and says nothing about costs.

Is the January effect real?

It is an old and widely cited finding, but Zhang and Jacobsen (2013) found in 300+ years of UK returns that the January effect only emerges around 1830, and concluded that monthly seasonals are sample-specific. Whether it remains in recent data is not something our sources let us state.

Why do researchers doubt calendar anomalies?

Sullivan, Timmermann and White (2001) showed that once you account for the full universe of rules that could have been tested, calendar effects no longer remain significant. Calendar effects, they argued, are a striking example of a data-driven discovery. Long-history evidence from the UK points the same way for monthly patterns.

References

1. Ariel, R. A. (1987). “A monthly effect in stock returns.” Journal of Financial Economics, 18(1), 161–174. doi:10.1016/0304-405x(87)90066-3
2. Bouman, S. & Jacobsen, B. (2002). “The Halloween Indicator, ‘Sell in May and Go Away’: Another Puzzle.” American Economic Review, 92(5), 1618–1635. doi:10.1257/000282802762024683
3. French, K. R. (1980). “Stock returns and the weekend effect.” Journal of Financial Economics, 8(1), 55–69. doi:10.1016/0304-405x(80)90021-5
4. Jacobsen, B. & Zhang, C. Y. “The Halloween Indicator, ‘Sell in May and Go Away’: Everywhere and All the Time.” Working paper, SSRN 2154873 (later versions titled “an even bigger puzzle”). doi:10.2139/ssrn.2154873
5. Keim, D. B. (1983). “Size-related anomalies and stock return seasonality.” Journal of Financial Economics, 12(1), 13–32. doi:10.1016/0304-405x(83)90025-9
6. Lakonishok, J. & Smidt, S. (1988). “Are Seasonal Anomalies Real? A Ninety-Year Perspective.” The Review of Financial Studies, 1(4), 403–425. doi:10.1093/rfs/1.4.403
7. McLean, R. D. & Pontiff, J. (2016). “Does Academic Research Destroy Stock Return Predictability?” The Journal of Finance. doi:10.1111/jofi.12365
8. Rozeff, M. S. & Kinney, W. R. (1976). “Capital market seasonality: The case of stock returns.” Journal of Financial Economics, 3(4), 379–402. doi:10.1016/0304-405x(76)90028-3
9. Sullivan, R., Timmermann, A. & White, H. (2001). “Dangers of data mining: The case of calendar effects in stock returns.” Journal of Econometrics, 105(1), 249–286. doi:10.1016/s0304-4076(01)00077-x
10. Zhang, C. Y. & Jacobsen, B. (2013). “Are Monthly Seasonals Real? A Three Century Perspective.” Review of Finance, 17(5), 1743–1785. doi:10.1093/rof/rfs035