Market Seasonality — Does \"Sell in May\" Actually Work?

"Sell in May and go away" — sell in May and come back to the market in fall — is probably the oldest "strategy" in finance: it rhymes, it sounds wise, and it requires no thinking. Seasonality is a broader family of ideas like this: the January effect, the Santa Claus rally, the turn-of-the-month effect, "Tuesdays on the S&P," summer rallies in silver, US-session hours on bitcoin. QuantifiedStrategies tested around 30 strategies in this category — and the results are more interesting than the old saying alone: some patterns in the data really do exist, and the flagship example is silver, with an extra annual return of roughly +3.9%.

Before you rearrange your portfolio around a calendar, though, read to the end — because the main lesson of this article isn't any single pattern, it's the small-sample trap. An annual pattern tested on 20 years of data is N=20. Twenty observations. Most seasonal folklore stands (and falls) on that number.

Educational disclaimer: this material is for educational purposes only and is not investment advice. The backtest results cited are historical, come from specific periods and markets, and do not guarantee future results.

What Seasonality Is and Where It Would Come From

Seasonality is the market's statistical tendency to behave differently depending on the calendar. It shows up at several scales:

• Annual: better/worse months (sell in May, the January effect, summer rallies in agricultural commodities).
• Monthly: the turn-of-the-month effect — pension fund inflows and payday cash buy stocks around the month change.
• Weekly: differences between days of the week.
• Intraday: different behavior at different hours — on crypto, for example, around the US session open.

For seasonality to be more than a curiosity, it should have an economic mechanism: a real, repeatable reason why money flows at a specific point in the calendar. A physical demand cycle in commodities (harvests, heating season, jewelry demand), tax settlements, month-end fund inflows, year-end bonuses and rebalancing. A pattern with a mechanism has a chance of surviving; a pattern without one is usually a data artifact that disappears the moment you start trading it.

What the Numbers Say — Sell in May, Silver and Bitcoin by the Hour

Sell in May. In very long data series (stock indices, decades of history) the November–April months really did average noticeably more than May–October — this is one of the best-documented calendar patterns, known in the literature as the Halloween effect. But the devil is in the stability: there have been long stretches where the "worse" half of the year performed very well (the last decade of the tech bull market regularly broke this pattern), and the edge in the data has weakened since it became widely known. Mechanically "selling in May" also means taxes, costs and the risk of missing strong months — after those adjustments, little is left of the old saying.

Silver. The strongest single result from the QuantifiedStrategies tests: a seasonal strategy on silver returned roughly +3.9% a year on top of plain buy-and-hold, and on a risk-adjusted basis the difference was crushing — 26.9% vs. 7.3% for buy-and-hold, because the strategy was only in the market for a fraction of the year. Here at least there's the skeleton of a mechanism (the industrial/jewelry demand cycle for precious metals). But before you call it a grail: this is still an annual pattern — a few dozen observations, sensitive to a handful of the best years in the sample.

Bitcoin intraday. At the other end of the timescale: backtests of BTC's intraday seasonality showed that returns aren't evenly distributed across hours — historically, activity around the US session stood out. The upside: an intraday pattern has thousands of observations, so the statistics are more honest than annual patterns. The downside: intraday patterns are sensitive to microstructure changes (ETFs, institutional trading hours) and to costs — an edge measured in fractions of a percent disappears in commissions faster than it appears.

Fact-check note: all figures come from QuantifiedStrategies publications — verify at the source and reproduce on your own data before using them. It's also worth remembering that out of 30 tested seasonal strategies, the prettiest ones get published and cited; that selection mechanism alone inflates the impression of how well seasonality "works."

The Main Lesson: N=20, or Why the Calendar Fools You So Easily

Now for the most important part. With annual patterns, one full repetition of the pattern equals one year. Testing "sell in May" on 20 years of data gives you a sample of N=20. For comparison: an intraday strategy backtest can have 2,000 trades. At N=20:

• A handful of years drives the whole result. Remove the two best years from the sample and the "edge" often disappears. This is a test worth running always: check the result without the three best observations.
• The standard error of the mean is huge. A difference that looks "obvious" on a chart of monthly averages is often statistically indistinguishable from noise.
• Multiple comparisons finish the job. There are 12 months, 5 weekdays, and dozens of holidays and "effects." Test 30 seasonal ideas (exactly as many as QS did) and even on purely random data a few will come out "profitable." The ones that come out get published — that's classic survivorship bias in research, not just in funds.

That's why solid work with seasonality looks like this: first the mechanism (why would money flow at this particular point?), then a test on data the pattern hasn't "seen" (other decades, other markets with the same mechanism), and finally the question of costs. A pattern that clears all three filters is rare — and that's exactly why something like silver's seasonality earns attention while "Thursdays are up days" shouldn't.

How to Use Seasonality Sensibly — If at All

1. Treat seasonality as a filter, not a system. The healthiest use: seasonality adds to a decision made for other reasons — for example, a trend-following strategy or mean reversion traded more aggressively in a statistically favorable period and more cautiously in an unfavorable one. A standalone "because it's May" signal isn't enough.
2. Demand a mechanism. A physical demand cycle, taxes, fund flows — if you can't name whose money is flowing and why in a given period, assume it's an artifact.
3. Count the sample honestly. N = the number of times the pattern repeats, not the number of candles. 20 years of an annual pattern is N=20; scale your conclusions to that number, with humility.
4. Test robustness. The result without the best 2–3 observations; split the sample in half (do both halves show the same thing?); the same pattern on a related market.
5. Don't rebuild your portfolio around the calendar. For a long-term investor, systematic dollar-cost averaging (DCA) without trying to time exits for summer has historically been hard to beat after costs and taxes. Seasonality is a garnish for an active trader who understands the statistics — for example, when timing entry into a swing position — not a reason to sell off a portfolio every May.

A reasoning example (illustrative): a trader is considering a long position in silver based on a trend signal. They check seasonality: it's entering a period that's historically favorable (demand mechanism + the QS result). Seasonality doesn't create the trade here — it raises its priority in the idea queue and lets the trader take the full, planned position size instead of half. If the calendar were unfavorable, the trade could still happen — just smaller, or with a higher signal-quality threshold.

When It Doesn't Work and the Most Common Traps

• Patterns with no mechanism. "Wednesdays on pair X are bearish" — if there's no money flow behind it, there's noise behind it. Noise doesn't pay.
• Self-inflicted data mining. Scanning data for 50 patterns and trading the ones that "worked out" guarantees a portfolio of artifacts. The more tests you run, the higher the bar of proof should be.
• The best years make the result. Commodity seasonality can be driven entirely by 2–3 years of supply shocks. Check the result after removing them.
• A known pattern is a traded pattern. Seasonalities described in every book (the January effect!) tend to fade or shift over time — the market arbitrages the calendar just like any other public information.
• Costs and taxes. A "exit in May, return in November" strategy means two taxable events and two full sets of costs every year. With an edge measured in single-digit percent annually, that's often the difference between profit and loss versus simply holding.
• Carrying patterns between markets. Natural gas seasonality doesn't imply bitcoin seasonality. Every market gets its own mechanism, its own test.

The lesson from seasonality is universal: the market has rhythms, but the calendar is the weakest kind of evidence, because it gives you the fewest observations. Before you believe any annual pattern, ask three questions: what's the mechanism, what's the N, and what's left after removing the best years. "Sell in May" usually doesn't survive those questions as a system — it survives as a reminder that exposure is worth adjusting to conditions, not holding rigidly fixed. And that's the most honest conclusion to draw from it.

FAQ

Rafał — Krypto Bez Ściemy

Trader and founder of Strefa Tradingu. He’s been breaking crypto down on YouTube for years — no hype, no signals, with a focus on market structure and risk management.

kryptobezsciemy.plInstagramYouTubeKlub Zero