“Sell in May” cut risk and return by about the same amount
Since 1993, stepping out of SPY every May cut its worst fall from 51% to 30%, but earned 8.1% a year instead of 10.8%. Return per unit of risk was statistically the same.
- Out May–Oct, since 1993
- 8.1%/yr
- Worst fall, out
- −30%
- Worst fall, in
- −51%
- Sharpe, out
- 0.56
- Sharpe, in
- 0.60
What we found
Since 1950, November–April beat May–October in 72% of years, and September is the only month with a negative average (−0.63%). Once all twelve months are tested, that isn’t significant (p = 0.008 against a 0.004 threshold).
Since 1993, with SPY’s dividends, being out May–October in T-bills earned 8.1% a year against 10.8% for staying in, with volatility of 10.4% against 14.7% and a worst fall of 30% against 51%. The Sharpe ratios were 0.56 and 0.60; a year-block bootstrap puts the difference between −0.31 and +0.23, so risk-adjusted they are the same.
The full results
| Measure | Always invested | Out May–Oct, in T-bills |
|---|---|---|
| Return a year | 10.8% | 8.1% |
| $10,000 became | $314k | $138k |
| Volatility | 14.7% | 10.4% |
| Worst fall | −51% | −30% |
| Sharpe ratio | 0.60 | 0.56 |
| Time in the market | 100% | 50% |
Sharpe difference −0.04; a year-block bootstrap puts it between −0.31 and +0.23.
| Month | Average | Median | Up in | 95% range of the average |
|---|---|---|---|---|
| January | +1.08% | +1.57% | 61% | +0.01% to +2.15% |
| February | −0.03% | +0.31% | 54% | −0.84% to +0.78% |
| March | +1.04% | +1.43% | 64% | +0.21% to +1.87% |
| April | +1.43% | +1.18% | 70% | +0.52% to +2.33% |
| May | +0.38% | +1.06% | 62% | −0.45% to +1.20% |
| June | +0.20% | +0.08% | 55% | −0.61% to +1.02% |
| July | +1.31% | +1.33% | 61% | +0.43% to +2.19% |
| August | +0.04% | +0.86% | 55% | −0.96% to +1.05% |
| September | −0.63% | −0.30% | 45% | −1.61% to +0.36% |
| October | +0.90% | +1.03% | 59% | −0.33% to +2.12% |
| November | +1.86% | +2.03% | 70% | +0.90% to +2.82% |
| December | +1.41% | +1.41% | 72% | +0.66% to +2.17% |
Do the months differ at all? p = 0.005 (Kruskal–Wallis) across all twelve. Prices only, no dividends.
- Nov–Apr
- May–Oct
| Decade | Nov–Apr | May–Oct |
|---|---|---|
| 1950s | +9.4% | +3.6% |
| 1960s | +6.8% | −0.7% |
| 1970s | +4.4% | −2.7% |
| 1980s | +9.0% | +4.3% |
| 1990s | +10.6% | +4.4% |
| 2000s | +0.2% | −1.5% |
| 2010s | +8.8% | +2.6% |
| 2020s | +6.5% | +8.8% |
Average six-month price return. Nov–Apr beat May–Oct in 72% of 75 years, but May–Oct still rose in 65%.
Is the seasonal pattern real?
Over 75 years, November–April beat May–October in 72% of years, and the gap is significant on its own. But September is the only month with a negative average, and once all twelve months are tested together the September result alone isn’t significant. November–April came out ahead in every decade except the 2020s so far.
The trade-off
Stepping out every May halves the time in the market. That gave up about 2.7 points of return a year since 1993, while cutting volatility by about 30% and the worst fall by about 40%. Risk-adjusted, the two are statistically the same: the strategy is a lower-risk way to own stocks, not a better one. Taxes on selling every year would tilt it further against switching.
How it was measured
- SPY with dividends from 1993; out of the market, cash earns the 3-month T-bill rate (FRED TB3MS).
- Sharpe ratios from monthly returns over the T-bill rate. No costs or taxes.
Caveats
- On S&P 500 prices alone (no dividends), the switch edges ahead from 1950 (8.7% against 8.3% a year). That comparison ignores the dividends it gives up, so it isn’t the one charted.
Questions
- Does “sell in May and go away” work?
- It lowered risk and return by about the same amount. Since 1993, out May–October earned 8.1% a year against 10.8% for staying in SPY, with a worst fall of 30% against 51%, and a statistically equal Sharpe ratio.
- What is the worst month for stocks?
- September, the only month with a negative average S&P 500 return since 1950 (−0.63%), though that alone isn’t statistically significant once all twelve months are tested.
More research
- Data studyBacktest on today’s biggest stocks and you double your return
- Market factThe 200-day rule sat out all 10 best and all 10 worst days of the last 20 years
- Data studyInsider buying, tested on 17,345 purchases: it didn’t beat similar stocks
Historical analysis. Past performance does not guarantee future results. Not investment advice. Published Nov 2026. Data: SPY and FRED to Nov 6, 2026.