Most trading seasonality is noise. Calendar patterns such as “sell in May” or weak Mondays show up in historical averages, but once you account for how many patterns were searched to find them, most stop being statistically meaningful. The calendar effects worth a day trader’s attention are the ones that change liquidity on known dates.
That distinction, between patterns in returns and patterns in market structure, is the whole answer. The first kind is mostly a statistical illusion. The second kind is scheduled, has an obvious cause, and you can plan around it.
What traders mean by seasonality
“Seasonality” covers two very different claims that usually get mixed together:
- Calendar return effects. The claim that prices tend to go up or down at certain times: the January effect, “sell in May and go away”, the turn-of-the-month effect, weak Mondays, pre-holiday strength, the Santa Claus rally. These are predictions about direction.
- Calendar structure. The fact that some dates and times reliably bring different participants and different volume: the open and close of each session, index rebalance days, option expiration, shortened holiday sessions, scheduled central bank meetings. These say nothing about direction. They describe liquidity.
The first kind generates most of the headlines. The second kind generates most of the practical value.
The best evidence that seasonality is real
The strongest published case is Bouman and Jacobsen’s “The Halloween Indicator, ‘Sell in May and Go Away’: Another Puzzle”, published in the American Economic Review in 2002. Examining monthly stock returns across major world markets, they found returns from May to October were lower than returns from November to April in 36 of the 37 markets they studied. It is a serious paper in a top journal, and later work by Jacobsen and co-authors reported that the pattern continued in the years after publication.
It is also contested. Maberly and Pierce, writing in Econ Journal Watch in 2004, re-examined the US evidence and concluded the result was not robust to alternative specifications, with a small number of extreme months doing much of the work. That is the normal state of the seasonality literature: a finding, a rebuttal, and no settled answer.
The best evidence that most of it is noise
The deeper problem is not any single pattern but the process that finds them. Sullivan, Timmermann and White tackled it directly in a paper later published in the Journal of Econometrics as “Dangers of data mining: The case of calendar effects in stock returns”. In the working paper version they built a universe of 9,452 calendar rules — every day-of-week, week-of-month, month-of-year, turn-of-month and holiday variation they could construct — and ran them against Dow Jones Industrial Average data from 1897 to 1996.
The best rule they found, which stayed out of the market on Mondays, produced a mean annualised return of 8.66 percent against 4.63 percent for simply holding the index, with an individual p-value that looked overwhelming. Then they asked the right question: how likely is it that the best of 9,452 rules looks this good purely by chance? Once evaluated in the context of the full universe of rules searched, their conclusion was that calendar effects no longer remain significant.
Why even a real seasonal effect rarely helps a day trader
Suppose a monthly pattern is real. It still has three problems for anyone trading intraday:
- It is tiny relative to one day’s noise. A difference of a few percent spread across six months works out to a few hundredths of a percent per session — far smaller than the ordinary daily range, and smaller than your trading costs.
- It has very few observations. A six-month pattern produces one data point per year. Thirty years of data is thirty observations, which is not enough to separate a real effect from luck with any confidence.
- It can fail for years at a time. An average across decades includes long stretches where it did the opposite. A trader relying on it may never live long enough, in account terms, to collect the average.
The calendar effects that do matter, and why
The dates worth marking are the ones where the reason is structural and obvious. They do not tell you which way price goes. They tell you when the market will be different.
| Calendar event | What actually changes | Why it happens |
|---|---|---|
| Time of day | Volume is heavy at the open and close, thin in the middle | Information arrives before the open; benchmarks are set at the close. See why volume dies at lunch |
| Scheduled data and central bank days | Liquidity drains before the release, then the range expands | Nobody wants to be filled just before a number they cannot see |
| Index rebalance and option expiration days | Unusually heavy trading into the close | Index funds must trade at the closing price on set dates |
| Earnings season | More single-stock gaps and wider ranges | Company news arrives outside the regular session |
| Holiday and half-day sessions | Thin books, erratic moves, early closes | Many participants are simply absent |
Each of these belongs in a pre-market check, and none of them requires believing that a month or weekday has a mood. They are covered in more depth in how economic releases move markets and earnings gaps.
How to test a seasonal claim before you believe it
- How many patterns were tried to find this one? If the answer is “we scanned everything”, discount it heavily.
- Is there a mechanism? “Funds rebalance at quarter-end” is a mechanism. “September is bad” is not.
- Did it work after it was discovered? Out-of-sample performance is the only kind that counts. Forward-test before you commit money; see forward testing a strategy.
- Is it bigger than your costs? An edge smaller than the spread and commission is not an edge.
- How often did it fail? Look at the individual years, not the average.
Frequently Asked Questions
Does “sell in May and go away” actually work?
There is published evidence that stock returns have been lower from May to October than from November to April across many markets, but later researchers have disputed how robust it is for US stocks. Even where the pattern exists, it is an average over decades, with many individual years going the other way. It is not a reliable rule for any single year.
Are some days of the week better to trade than others?
Day-of-week effects, such as weak Mondays, appear in older data but are among the patterns that stop being significant once researchers account for how many calendar rules were searched. Differences between weekdays in volume and scheduled data releases are real; differences in expected return are not something to build a plan on.
Why do seasonal patterns stop working after they are published?
Two reasons. Some were never real in the first place and were found by searching many possible patterns until one looked good. Others may have been real but small, and once enough traders act on them, the trading itself removes the edge.
Should a day trader use seasonality at all?
Only the parts with a clear mechanism: the intraday volume curve, scheduled data releases, index rebalances, option expirations and shortened holiday sessions. These change liquidity on known dates. Calendar return patterns are too small and too inconsistent to decide an intraday trade.
Bottom line
Seasonality splits into two claims, and only one of them survives scrutiny. Calendar return patterns can be found in any long data set, but when Sullivan, Timmermann and White tested 9,452 of them on a century of Dow data, none remained significant once the size of the search was counted — and even the best-supported one, the Halloween effect, is still argued over. Calendar structure is different: the open, the close, data releases, rebalances and holidays change who is trading and how deep the book is, for reasons anyone can explain. Plan around the structure and ignore the folklore. For the mechanics underneath all of it, start with how markets actually work, and see our FAQ for how the room prepares for scheduled events.