The worst month for stocks, and why September keeps winning that title
August 9, 2026 · 7 min read
Ask which month is worst for stocks and you will get one answer: September. It is the only month with a negative average return across the long US record, and it has held that position across most sub-periods you can slice it into.
That much is settled. Everything after it is weaker than it sounds.
Every month, ranked
The numbers below are computed from the S&P 500 (^GSPC) daily closes, January 1990 to August 2026, on total price return excluding dividends. Every figure is reproducible from that series.
| Month | Average | Median | Positive | Worst |
|---|---|---|---|---|
| Jan | +0.62% | +1.50% | 61% | -8.6% |
| Feb | +0.04% | +0.69% | 57% | -11.0% |
| Mar | +0.82% | +1.11% | 62% | -12.5% |
| Apr | +1.74% | +1.22% | 73% | -8.8% |
| May | +1.25% | +1.24% | 76% | -8.2% |
| Jun | +0.10% | +0.09% | 59% | -8.6% |
| Jul | +1.42% | +1.62% | 62% | -7.9% |
| Aug | -0.49% | +0.36% | 56% | -14.6% |
| Sep | -0.72% | +0.15% | 50% | -11.0% |
| Oct | +1.46% | +1.87% | 64% | -16.9% |
| Nov | +2.17% | +2.81% | 75% | -8.0% |
| Dec | +1.24% | +1.12% | 72% | -9.2% |
The number that undoes the headline
September averages -0.72% and is the only negative month on the board. Its median is +0.15% — positive.
Those two facts together are the whole story. Half of all Septembers since 1990 finished up. The negative average is produced by a small number of severe ones, not by Septembers generally being bad. Its win rate is exactly 50%: a coin flip.
Compare the two charts. The mean chart says avoid September. The median chart says September is merely unremarkable. Every article you have read about the worst month for stocks used the first one.
What the ranking actually looks like
The month-by-month ranking has three features worth knowing:
- September is negative on average (-0.72%), and August is the only other negative month (-0.49%).
- The spread from best to worst is 2.9 points — November (+2.17%) to September (-0.72%). Meaningful, but smaller than a single bad week in a normal year.
- The ordering of the middle months is unstable. Which month sits fourth versus eighth changes depending on the start year you pick, which is a sign those differences are noise.
Only the extremes survive re-slicing. The middle of the table is an artefact of the sample.
The explanations, and why to distrust them
Several stories get offered for September. Funds rebalance before fiscal year end. Traders return from holiday and reassess. Tax-loss selling starts early. Investors de-risk ahead of an historically volatile October.
Each is plausible. None is testable in a way that would distinguish it from the others, and the effect predates most of the institutional structures the explanations depend on. When an anomaly is older than every proposed cause for it, the honest position is that the cause is unknown.
That matters practically. A pattern with an understood mechanism can be trusted to persist while the mechanism does. A pattern without one can stop at any time, and you will not know why or when.
What a weak month does not mean
A negative average does not mean most Septembers are down. 50% of them were up, the median was +0.15%, and the worst on record was -11.0% — that single month is doing most of the work in the average. If you sat out every September you would have avoided a few disasters and missed a larger number of ordinary, mildly positive months.
This is the recurring shape of seasonal statistics: the average is driven by the tail, and the tail is what you cannot time.
Build this table for your own ticker
- Pick the ticker in the symbol selector and set a 20-year range.
- Open the Intramonth tab for the month-by-month breakdown of that instrument.
- For any month that looks weak, switch to Trading Analysis, set the period to that month, and read the median beside the average.
- Where the two disagree sharply, the average is being set by one or two years — check the worst year to see which.
Dispersion between individual stocks is far larger than dispersion between months of the index, which is why a per-ticker table is worth building and a market-wide one is trivia.
Open TradeSeasonsThe version of this that is usable
Ranking months for "the market" is a parlour game. Ranking them for a specific instrument is research, because the dispersion between individual names is far larger than the dispersion between months of the index, and the seasonal shape of a single stock is often driven by something concrete: a product cycle, a harvest, a fiscal calendar, a weather-dependent input cost.
That is where a monthly breakdown starts being worth acting on — one instrument, its own history, with the win rate and the worst year sitting next to the average.
This article describes historical price behaviour and is for information only. It is not investment advice, and nothing here is a recommendation to buy or sell any security. Past performance does not predict future returns.
Check any of this against 30 years of history
Every pattern in this article is one you can reproduce on your own tickers in a couple of clicks.
Start Analyzing