Two kinds of seasonality
Seasonal patterns in markets split into two groups, and only one of them deserves your attention:
Patterns with a mechanism
- Month-end and quarter-end rebalancing — large funds mechanically adjust allocations. The flows are real and reasonably predictable in direction, though the size varies.
- Holiday liquidity — thin markets around major holidays produce wider spreads and exaggerated moves. A mechanical consequence of fewer participants.
- Data-release clustering — certain days of the month reliably carry specific releases, so volatility clusters there. This is just the calendar.
- Tax and fiscal deadlines — real money moves for real reasons on known dates.
Patterns without a mechanism
- "January effect" in its original small-cap form — largely arbitraged away once widely known.
- "Sell in May" — works in some samples, fails in others, with no coherent cause.
- Day-of-week effects in major FX — mostly disappeared once measured properly.
How seasonality studies fool you
If you test twenty seasonal patterns on the same data, one will look statistically significant by chance. This is the multiple-comparisons problem, and it is why published seasonal tables are far less reliable than they appear. Any pattern you did not predict before looking at the data is a hypothesis, not a finding.
Testing honestly
- State the hypothesis and the mechanism before you test.
- Split the sample: find the pattern in one half, confirm it in the other.
- Check whether it survives transaction costs — seasonal edges are usually small.
- Ask what changed: if a pattern worked for twenty years and stopped, something structural changed.
The honest caveat
Genuine seasonal effects tend to be small relative to noise and get arbitraged away once widely known. Treat seasonality as a tiebreaker or a risk-timing consideration, never as a primary thesis.
How to check this yourself
Take any price series and compute the average return by day of the week. You will find a "pattern" — usually a weak Monday or a strong Friday. Now ask what mechanism would cause it. If you cannot name one, you are looking at noise dressed up as a pattern, and the same test applies to every seasonal chart you have ever been shown.