Sell in May and go away: does it still work?

Sell in May and go away means holding stocks from November to April and sitting in cash from May to October. In the US, this earned less than simply holding the market. A $10,000 investment at the start of 1999 grew to about $58,000 with the rule, against about $102,000 for buy-and-hold.

Sven Bouman and Ben Jacobsen made the rule famous in 2002. Their paper found that the winter half beat the summer half in 36 of 37 countries, using data from 1970 to 1998. The saying is much older: London traders used to sell in May and come back on St Leger Day in September.

We asked whether the pattern survived after that paper, and whether anyone could have made money from it.

How we tested it

The rule

Winter means November to April. Summer means May to October. The rule sells at the close on the last trading day of April and buys back at the close on the last trading day of October. The calendar is known in advance, so the rule never uses information it could not have had. That is about two trades a year.

Every switch costs 0.15%. Money out of stocks earns the Treasury bill rate, 3.35% a year on average. Leveraged versions pay the bill rate plus 0.4% to borrow, plus a 0.95% yearly fee. We ignore taxes.

The data

For the US we used the S&P 500 with dividends reinvested, daily, from 1928 to October 2026. Before 1993 dividends are estimated from Robert Shiller’s yield data; from 1993 we use the SPY fund. Cash returns come from Ibbotson’s Treasury bill series until 1933, the Federal Reserve’s 3-month bill rate until 1959, and the 13-week bill yield after that.

For earlier years we also checked Shiller’s monthly series, which starts in 1871. It uses monthly average prices, so it is blurry around month ends.

Outside the US we used 41 data series covering 23 countries: 20 stock indexes (four of them US indexes) and 21 country ETFs. The indexes are price indexes in local currency, so they exclude dividends. European companies pay most dividends in spring, which can make the winter–summer gap look bigger than it is. The ETFs include dividends and are priced in US dollars. They start in 1996.

Three periods

We split the US history into three parts and read each one separately.

• 1928–1969, before the paper’s data. A real pattern should show up here too.
• 1970–1998, the years the 2002 paper used.
• 1999–2026, after the paper. This is a clean test, because nobody could have fitted the rule to these years.

How we checked for luck

A few huge crashes can make a seasonal rule look brilliant. So we ran four checks on the main results:

• Drop the best years. We removed the 1, 3 and 5 years that helped the rule most, then measured again.
• Count years, not markets. Markets fall together, so 41 series in the same year are not 41 separate results. We averaged across markets each year and treated each year as one result.
• Shift the calendar. We moved the six-month “winter” window to each of the 12 possible starting months. A real effect should rank near the top among those 12 windows.
• Same exposure, no timing. The rule holds stocks only half the year. We compared it with a fixed mix that holds the same average amount of stock all year, with no timing.

We also resampled whole years at random to see how wide the range of luck was. When a result fell inside that range, we say so in the text.

The pattern survived abroad, but it rests on a few years

From 1999, winter beat summer in 40 of the 41 series. Because markets crash together, as almost every country did in autumn 2008, 41 series are not 41 independent results. The real sample is 28 years.

To see whether a few lucky years did the work, we dropped the best ones. About half the gap came from three summers with sharp falls: 2002, 2008 and 2011. Without them the evidence went from borderline to indistinguishable from chance.

In the US it barely existed before 1970

From 1929 to 1969 the US winter–summer gap was about 0.7 points a year. That is too small to tell apart from normal year-to-year noise. In the paper’s period, 1970 to 1998, it was 5.0 points. Shiller’s data from 1872 to 1969 also shows almost no gap.

A pattern that is missing before the years in which it was discovered is a warning sign. From 1999 the US gap came back at about 3.4 points a year (counting full summers only, through 2025), but without the five best years it turned slightly negative.

Cash in summer cost more than it saved

*Largest fall from a peak to a later low.

Selling for the summer cut the worst drop from −55% to −36%. But it grew about 2.2 points a year slower on average, and over 28 years that became the gap between $58,000 and $102,000.

Summer was weaker than winter, but it was not bad. The US summer half still earned about 1.6% more than cash per year. Selling in May meant giving that up every year to dodge crashes that came only now and then.

The season does carry some information. After 1970 the rule beat a fixed mix of about half stocks and half cash, which has the same average exposure. Before 1970 it lost even that comparison. It never beat simply holding the market.

None of the wins in this table are large enough to rule out luck.

Leverage in winter: more growth, much deeper crashes

We tried borrowing to hold twice the stock position in winter. With a normal position in summer, it grew 10.6% a year, more than buy-and-hold. But its worst drop was −74.9%.

The problem is timing. The last leg of the 2008–09 bear market, from November 2008 to March 2009, fell in the winter half. So did the Covid crash of February and March 2020.

We also tried four real leveraged funds held only in winter. Only the 3x S&P 500 fund, UPRO, beat its plain index, at 18.3% a year against 15.2% from mid-2009. Its worst drop was −76.8%, no smaller than holding the 3x fund all year. The 2x S&P 500 fund and both Nasdaq funds lost to their plain index.

What about September?

From 1929 to 2026, September was the weakest month for the S&P 500 in each of the three periods we checked. In Shiller’s data from 1872 to 1927, though, it was one of the better months: 10th worst of 12. Since 1999 September has also been weaker than other months in all 40 index and ETF series we had, but those cover the same years as the US data, so they are not separate evidence.

Skipping only September grew 10.0% a year against 8.7% from 1999 to 2026. Moving the switch by a single day cut that to 9.5%, so much of the edge depends on the first trading day of October. It also edged out buy-and-hold in 1929–1969 and 1970–1998. But the 1999 result was within the range of luck, and removing the three worst Septembers (2002, 2008 and 2022) cut the advantage to almost nothing.

There is also a selection problem: we picked September because we already knew it was the worst month. When we let the data pick the worst month each year using only past years, the rule lost to buy-and-hold in 1950–1969 and in 1970–1998.

Other markets and crypto

An equal-weight basket of 21 country ETFs that sat in cash every summer did beat holding the basket from 1999: 9.1% a year against 8.2%, with a worst drop of −38% against −61%. But 2008 did most of the work. From 2009 the rule grew 7.9% a year against 9.6%.

For Bitcoin (2011–2026), holding only from November to April grew about 51% a year, against about 81% for holding all year. The winter gap came almost entirely from two early years, 2011 and 2013. For Ethereum (2017–2026), the gap vanished in the second half of the data. Since 2022 the worst month for both has been June, not September.

What if you are waiting with spare cash?

Some people get money in summer and plan to wait until November to invest it. Compared three years later, waiting left them behind in about 6 of 10 cases, by about 1% at the median.

Money that arrived from mid-August to September and waited until the end of September did no harm. It also gained almost nothing: +0.2% at the median.

Does it help inside an uptrend?

We split S&P 500 days by season and by whether the index was above its 200-day average, the average closing price of the last 200 trading days. Winter days in an uptrend had the best return for the risk taken in every period.

A rule built on that, holding more in winter uptrends and going to cash in downtrends, added about one point a year from the season. But the trend filter cost more than that after 2009, and the rule fell −43% in 2022–23, when buy-and-hold fell −24.5%.

What this means for you

• Don’t move to cash in May. In the US it cost about 2 points a year.
• If money arrives in summer, invest it now. Waiting for November usually left you behind.
• If money arrives in late August or September, waiting until the end of the month did no harm, but it gained almost nothing.

This test ignores taxes. Why a seasonal pattern might exist is still debated. Bouman and Jacobsen pointed to summer holidays. Other researchers have suggested that investors take less risk when the days get shorter.

Bottom line

Winter has beaten summer in many countries, but much of that comes from a handful of summer crashes, and in the US the pattern barely existed before 1970. Selling in May cut the worst losses, but it earned less than holding the market, because the summer half still paid you to stay invested.