“Sell in May” Is Wrong — But the Cycle Is Turning Down

The Saying That Stopped Working

“Sell in May and go away” is one of the most quoted seasonal adages in markets. The folklore version says investors should exit equities at the start of May and stay out until November, on the theory that the May-through-October stretch historically underperforms. For the broad May-to-October window, there is still a measurable seasonal drag — it is just much smaller than the saying implies.

But inside that six-month window are pockets of strong upside. The most consistent one in recent years is the late-May-to-early-June stretch on the S&P 500. Specifically, the 20-day window from May 20 to June 9 has been a long trader's dream: 10 winning trades out of 10 over the last decade, with an average return of +2.5% and a max drawdown of just 1.0% on the worst trade.

If you had sold in May the way the adage suggests, you would have missed every one of those gains. Across the full lookback to 1950, the pattern has run 76 trades with a 56.6% win rate and compounded $1,000 into $1,324. The recent decade is the strongest stretch inside that history.

The Trade: Long May 20 → June 9

The mechanical version of the pattern is straightforward. Each year, buy SPX at the close on or near May 20 and exit at the close on or near June 9. No filters, no stops, no discretion. Here is what that produced from 2016 through 2025.

Every year of the past decade has been a winner. The smallest gain was +0.99% in 2024; the largest was +7.93% in 2020, when the trade captured the back end of the COVID rebound. The worst intraday drawdown across all ten trades was just 1.0% — meaning the pattern has not only been profitable, it has been calm.

The Catch: The Equity Curve Cycle Is Turning Down

Here is where most seasonal analysis stops — with a long string of wins and a recommendation to take the trade. The problem is that even strong patterns have a rhythm. They run hot for a while, then cold, then hot again. The recent decade has been the hot half of that rhythm. The question worth asking is: what does the longer pattern say about right now?

We can answer that by compounding the full history of the pattern into an equity curve and running spectral analysis on the result. The compounded curve from 1950 through 2025 contains 76 trades and reaches $1,324 from a $1,000 start — a 32.5% total return with a 56.6% win rate. Crucially, that curve does not climb in a straight line. It alternates between hot streaks (long runs of green) and cold streaks (clusters of losses), with a dominant cycle period of roughly 16 trades.

The Bartels test — a statistical check for genuine cyclicality — sits at p = 0.314 in the latest reading, which does not clear the conventional significance threshold. That is itself informative: after a decade of unbroken wins that pushed the equity curve well above its long-term mean, the rhythm of hot and cold phases has become less periodic and more skewed. The visual character of the curve still shows clear cold-phase clusters — what has changed is that the recent run is so far above prior norms that it dilutes the cycle's statistical signature.

This is the same insight that drove our earlier piece on equity curve regime filtering: a high win rate tells you the pattern works on average, but cycle phase tells you whether it is working now. Mature traders use both.

What Has Happened When the Cycle Was Going Down

If recent results were all the data we had, there would be no reason to worry. The discomfort comes from looking at the full 75-year record. The pattern's lifetime win rate is 56.6%, not 100%. That number is the average across both halves of the cycle — and the gap between the recent decade (100%) and the lifetime average has to be filled in somewhere. It is filled in by the cold phases.

Pulling the recent 10 wins out of the dataset leaves roughly 66 trades from 1950 through 2015. Those trades produced approximately 33 wins and 33 losses — almost exactly a 50% win rate. In other words, before 2016 the pattern was a coin flip. The recent decade is the outlier, not the rule.

Three cold phases visible on the equity curve

The compounded curve from 1950 to 2025 makes the cold phases easy to spot. They are the long flat stretches and the multi-trade drawdowns where the orange cycle line is below the equity dots. Three of them stand out.

The shape of a cold-phase trade

Cold-phase losses are usually not dramatic. Looking through the 35 historical losing trades, the average losing trade comes in around –2% to –3%, with a handful of worse outliers reaching –5% to –7%. The pattern does not blow up — it simply stops working. Hot-phase wins of +2% to +8% get replaced by chop: a small win, a small loss, another small loss, a marginal win.

The lifetime maximum drawdown of −17.1% on the compounded curve was not produced by a single catastrophic year. It accumulated trade by trade, mostly during the 1950s–1970s structural cold phase. That is the characteristic signature of this pattern's cold side: quiet, repeated underperformance rather than a single bad event. The risk during a cold phase is not a crash — it is the slow grind of small losses that prevent compounding for years.

What the Filters Say

We ran two independent filters on the recent 10-trade window to see whether either would have skipped any of the wins. The first is the Williams Accumulation/Distribution indicator: only take longs when WAD is above its 57-period moving average. The second is the equity curve HOT regime filter: only take longs when the cycle phase is hot.

The Williams A/D filter passed 9 of the 10 trades and kept the 100% win rate. It did not flag this pattern as risky during the run. The HOT regime filter, in contrast, was far more selective — it only took 3 of the 10 trades. Those three still won, but a run filtered by the HOT signal captured far less of the decade's gains.

The two filters reveal different things. In a hot phase the unfiltered pattern tends to run cleanly, so even loose confirmation tracks the outcome. In a cooling phase, the HOT regime filter becomes more discriminating — precisely because the unfiltered pattern is statistically more likely to break. The current cycle reading is moving toward the second kind of environment.

The 2026 Setup: What the Data Is Showing

The next instance of this pattern runs from May 20, 2026 to June 9, 2026. Compared with the last ten years, the 2026 setup is unusual: every standard read on the pattern itself is strong, but every long-horizon read on the cycle is weakening at the same time.

The point of running both reads — recent base rate and long-term cycle — is that they can disagree, and that disagreement is itself information. “Sell in May” would have looked wrong for ten straight years. The cycle framework says the next ten may not look the same.

Frequently Asked Questions