“Sell in May and go away” is the most durable piece of market folklore in circulation, and unlike most folklore it has statistical support. Equity returns in the November-to-April half of the year have historically exceeded those in the May-to-October half, across many markets and over long samples.
The pattern is real in the data. Whether it is usable is a different question, and the answer is mostly no — for reasons that illustrate a general problem with seasonal strategies.
What the evidence shows
The finding has been documented across dozens of national markets and in samples stretching back to the nineteenth century in some cases. The winter half-year has outperformed the summer half-year on average, and the gap is large enough that it is unlikely to be pure chance.
That is a genuinely surprising result and it has survived a lot of scrutiny. It is not the same as being tradable.
The problems
The average conceals enormous variation. The summer half-year is positive in most individual years. It is the average that is lower, driven substantially by a modest number of severe episodes — several of the worst market events on record happened to fall in the autumn. Remove a handful of observations and much of the effect weakens considerably. A strategy whose edge depends on a few extreme events is not a reliable annual rule.
Sitting out costs money. Being out of the market for six months a year means forgoing dividends and, in most years, positive price returns. The alternative — holding cash — earns something, but the comparison depends heavily on the prevailing interest rate. The strategy looks far better in a high-rate environment than a low-rate one, which means its historical record partly reflects the rate environment of the sample rather than a seasonal effect.
Transaction costs and tax. Two round trips a year in a taxable account converts long-term holdings into short-term gains, and the tax differential alone can exceed the entire measured seasonal edge.
Publication. The effect was documented decades ago and is now among the best-known anomalies in finance. Patterns that are widely known and easily traded tend to attenuate. There is evidence that this one has weakened since it was popularised, which is what theory predicts should happen.
The explanations, and why they are unsatisfying
A statistical pattern without a mechanism is a pattern waiting to disappear. Several mechanisms have been proposed and none is fully convincing.
- Holiday liquidity. Institutional participation falls during northern hemisphere summer, thinning liquidity and raising risk premia. Plausible, but it predicts higher volatility rather than lower returns, and the effect appears in southern hemisphere markets too.
- Seasonal affective patterns. Research has linked risk appetite to daylight hours. Suggestive, and it does accommodate the hemisphere problem, but the effect sizes required are large.
- Bonus and allocation cycles. Year-end bonuses and new-year allocations concentrate inflows in the winter half. This has real support and is probably part of the story.
- Data mining. The honest null hypothesis. Researchers have tested an enormous number of calendar patterns; some will appear significant by chance. The May–October split survives out-of-sample testing better than most, which is a point in its favour, but not a decisive one.
The general lesson
This is a useful case study because it is among the strongest seasonal effects in the literature, and it still fails the practical test. If the best-documented calendar anomaly cannot reliably be traded after costs, the weaker ones — day-of-week effects, month-end effects, the various presidential and election-cycle patterns — deserve considerably more scepticism than they receive.
The questions to ask about any seasonal claim:
- How many independent observations? Ninety years of data is ninety observations of an annual pattern, not 22,000 daily ones. That is a small sample.
- Does it survive removing the extremes? If three years drive the result, it is not a rule.
- Is there a mechanism that predicts it in advance? An explanation constructed after the pattern was found is not evidence.
- What are the costs? Measured against turnover, tax and the return forgone while out of the market.
What to actually do in May
Nothing, on account of the calendar.
If the summer months genuinely make you uncomfortable, the honest interpretation is that your allocation is too aggressive for your risk tolerance year-round, and the correct response is to change the allocation permanently rather than to trade around a date. An investor who needs to be out of the market for half the year to sleep is not describing a seasonal insight. They are describing a portfolio that does not suit them.
The broader arithmetic — that costs and turnover reliably erode returns while forecasting reliably fails to add them — is the same one that underlies the case for low-cost, low-turnover investing generally, which we set out in index funds versus active management.
