Summer vs. Fall vs. Winter: What Stocks Actually Did the Past Five Years
“Sell in May and go away.” “Watch out for October.” “Santa Claus rally.” We tested the sayings against 155 years of S&P 500 data. Over the last five years, the single most reliable season was autumn — and the worst was winter. That is close to the exact opposite of the folklore.
Read this as a history lesson, not a calendar. Everything below describes what already happened. Seasonal patterns are weak, unstable, and reverse without warning — as you are about to see them do. This is educational analysis, not a recommendation to buy or sell anything at any time of year.
The last five years, by season
Here is the S&P 500's average return in each season over its five most recent completed runs, next to the average since 1950.
| Season | Last 5 yrs | Up | Since 1950 | Up |
|---|---|---|---|---|
| Summer | +3.72% | 4/5 | +1.19% | 51/76 |
| Fall | +4.31% | 5/5 | +2.64% | 55/76 |
| Winter | -0.52% | 2/5 | +2.69% | 49/76 |
| Spring | +4.21% | 4/5 | +2.50% | 54/76 |
Two things jump out.
Fall was perfect. September through November rose in all five of the last five years, averaging +4.31%. For the season everyone is supposedly afraid of, that is a remarkable run.
Winter was the problem. December through February averaged -0.52% and rose in only 2 of 5 years — even though, since 1950, winter has been one of the stronger stretches at +2.69%. The season with the friendliest reputation has lately been the one that lost money.
“Sell in May” worked for decades — then stopped
The oldest seasonal saying in the business says to hold from November to April and sit out May to October. Over the long run, the numbers genuinely support it.
Since 1950, the November–April stretch averaged +5.97% while May–October averaged +3.05% — a gap of about 2.9 percentage points per year in favour of the winter half. That is a real, persistent historical edge, and it is why the saying survived.
Over the last five years it inverted. May–October averaged +9.16%; November–April averaged +3.55%. The summer half beat the winter half by roughly 5.6 points a year. An investor who dutifully sold every May for the past five years would have sat out the better half of the year, five times running.
This is the most useful thing in the whole article, and it has nothing to do with seasons. A pattern can be real in the data for seventy years and still stop paying the moment enough people act on it. That is roughly what “the market is efficient” means in practice.
October is the most libeled month in finance
October owns the scariest résumé in market history. The worst single month in this entire 155-year dataset is October 1929, at −26.47%. October 1987 and October 2008 are in the same hall of fame.
And yet, since 1950, October has been the second-best month of the year, averaging +1.35% and rising 67% of the time. Over the last five years it was the best month outright, averaging +3.34% and finishing higher all five times.
Both facts are true at once, and the reason is worth understanding. October is genuinely volatile — it has one of the widest spreads of outcomes of any month. Volatility means big moves in both directions. We remember the crashes because crashes are memorable, not because they are typical. The average October is quietly good; the famous Octobers are catastrophic. A handful of disasters shape the reputation of a month that has mostly paid.
The month that actually drags
If you want a genuinely weak month, it is not October. It is September — the only month with a negative average return since 1950, at -0.06%, and the lowest hit rate of any month at just 54% positive.
| Month | Last 5 yrs | Since 1950 | % up |
|---|---|---|---|
| January | +1.53% | +0.75% | 55% |
| February | -1.15% | +0.43% | 59% |
| March | +0.62% | +1.22% | 72% |
| April | +0.77% | +0.72% | 61% |
| May | +2.04% | +0.46% | 61% |
| June | +2.74% | +0.71% | 64% |
| July | +1.59% | +0.43% | 61% |
| August | -0.19% | +0.00% | 58% |
| September | -0.60% | -0.06% | 54% |
| October | +3.34% | +1.35% | 67% |
| November | +1.62% | +1.26% | 74% |
| December | +0.44% | +1.40% | 67% |
Even here, keep the size in perspective. September's long-run average is -0.06% — a rounding error, not a crash. The gap between the best and worst month since 1950 is under two percentage points, while a single average month swings around three to four points either side of its mean. The noise is roughly twice the signal.
Notice too how little the two columns agree. February is the worst month of the last five years (-1.15%) but is mildly positive since 1950. June rose in all five recent years. If seasonality were a dependable force, those columns would line up. They do not.
So what should a beginner take from this?
Three things, none of which involve a calendar.
The effects are tiny next to the noise. Seasonal differences run to one or two percentage points a year. Ordinary month-to-month variation is several times that. You cannot reliably harvest a signal smaller than the noise around it.
Patterns decay once they are famous. “Sell in May” is the cleanest example in this dataset: a real edge for decades, then a five-year reversal. Anything printed in enough newspapers gets traded away.
Time in the market beat timing the seasons. Every seasonal strategy here involves being out of stocks for part of the year. Over the last five years the market rose in four of five springs, four of five summers and five of five autumns. Sitting out any stretch of that mostly meant missing gains and paying taxes for the privilege.
The reason these sayings persist is that they are easy to remember and occasionally right — which is exactly what makes them dangerous. A rule that works most years teaches you to trust it right up until the year it does not.
Test a season yourself, risk-free
The simulator runs on real market prices with $10,000 of practice money. Buy something, leave it alone, and watch what a season actually feels like from the inside.
Start paper trading →How we calculated this
We used the monthly S&P 500 price series compiled by economist Robert Shiller of Yale, published as an open dataset and extended after mid-2023 with the Federal Reserve Bank of St. Louis's S&P 500 series. It runs from January 1871 to September 2026.12
Three caveats, stated plainly:
- These are monthly averages of daily closes, not month-end closing prices — that is how Shiller's series is built.1 It smooths the data slightly. It does not change which seasons come out ahead.
- Price only, no dividends. Total returns would be roughly 1.5 to 2 points a year higher across the board, spread fairly evenly, so the comparison between seasons holds.
- Seasons are defined as spring = March–May, summer = June–August, fall = September–November, winter = December–February (labelled by the year it ends). “Last five years” means the five most recent completed runs of each season as of October 2026; autumn 2026 is still in progress and is excluded.
The S&P 500 is a US large-company index. None of this describes bonds, international markets, crypto or any individual stock.
Sources
- Robert J. Shiller, online data for Irrational Exuberance, Yale University — monthly S&P 500 price, dividend and earnings series since 1871 — econ.yale.edu
- DataHub / Core Datasets, “S&P 500 Index Data” — tidied CSV of the Shiller series, extended with FRED data — github.com/datasets/s-and-p-500
- Federal Reserve Bank of St. Louis, S&P 500 series (SP500) — fred.stlouisfed.org