Does Shorting A Stock After A 20% Drop Work? A 13-Year Backtest
Published October 19, 2026 · 6 min read · Data: 2013–2025 out of sample, plus 2026 year to date
The Short Answer
Not reliably. Shorting stocks that fell 20% or more in a day made +1.87% per trade in 2026, which looked like an edge. Tested on the 13 years before that, with rules set in advance and realistic costs, the same trade averaged -0.33% and was not statistically different from zero. It did beat shorting random stocks, but both lost money.
On This Page
The Idea
After a stock falls 20% or more in one day, some traders expect the selling to continue and short the stock the next day. Our 2026 data suggested it paid. A result found in one year is exactly the kind that fades, so we tested it on 2013–2025, data that had played no part in finding it.
The Rules We Fixed Before Testing
- Signal: a one-day drop of at least 20% in a US stock priced at least $5.
- Entry: short at the open two days after the drop, not the day after. A next-day open is often impossible to get on a crashed stock.
- Exit: cover after five trading days, or sooner if the stock rises 15% against us (a stop). The average hold was 4.25 days.
- Costs: 0.60% slippage and a 30% annualised borrow fee, both deliberately harsh because crashed stocks are costly to short.
- Hedge and control: each trade is adjusted for market beta, and compared with the same rules applied to random tradeable stocks.
The Results
Bars are the average; whiskers are the 95% range. Gray bars are the same rules applied to random tradeable stocks.
| Period | Trades | Avg Per Trade | 95% Range | Random Control |
|---|---|---|---|---|
| 2013–2025 | 4,286 | -0.33% | -0.94% to +0.24% | -0.96% |
| 2026 so far | 540 | +1.87% | -0.12% to +3.87% | -1.40% |
- It beat the control, but that is a low bar. The edge over random shorts was +0.63% per trade (range +0.08% to +1.15%), so crashed stocks did fall more than random ones. After these costs the absolute result was still negative.
- A few trades decide the outcome. Excluding the best 1% of trades, the average falls to -0.91%. The worst single trade lost about 290%.
- The stop is hit often. 28.3% of trades reached the 15% stop.
It Is Inconsistent From Year To Year
Seven of 14 years were positive. The worst years were 2016, 2018, 2020 and 2022. 2026 is year to date.
Against its random control the trade did better, ahead in 9 of 14 years, but the control is also negative every year because shorting costs are high. Weak markets were worst: when SPY was below its 200-day average the trade averaged -1.38% (range -3.05% to +0.29%), against -0.06% when it was above.
What Moves The Result
Two splits look interesting, and both are cautionary. Bigger drops did better, but the biggest group is the smallest and the least certain.
Average net return per trade. Only the largest drops are positive, and that range crosses zero.
Average net return per trade by price. We found the $40+ result after seeing the data, so it is a hypothesis, not a rule.
Why Backtests Of Shorting Flatter The Results
- Borrow availability. The test assumes every stock can be shorted. Many crashed stocks cannot be borrowed at all, or only at fees far above 30%.
- Squeezes. Crashed small stocks can bounce violently. A stop helps, but gaps can blow through it. See What Is Short Selling?
- Daily prices. We cannot see whether the high or low came first, so fills are idealised.
- Survivorship. Names that went bankrupt are mostly missing from the data, which understates short profits, so this one cuts in the shorts' favour.
What We're Doing About It
We froze the rules above and started a forward test on October 7, 2026. Each signal day is logged before the outcome is known, and we will judge the result only after at least 100 resolved trades across at least 60 different signal dates. The rules will not be changed to improve the numbers. Until it passes, we do not treat shorting big drops as an edge, and nothing on stkmrkt should be read as a prediction.
Forward Test Progress
Collecting DataWe show no average return until both targets are met, so a few early trades cannot be read as a result. Signals appear on only some days, so this takes months.
Signals counted from October 7, 2026. Last checked October 23, 2026. Rules are frozen and will not change.
Limits And Disclaimer
How We Tested This, And What It Can't Tell You
- Only the main spec was pre-declared. The grid of entry rules, stops and cost levels, and every split by drop size, price and market regime, are exploratory.
- No position limits. Each signal is a separate trade with no cap on how many are open at once, so this is not a portfolio simulation.
- Before tax, simplified costs. Real borrow fees and slippage vary by stock and day.
- Not advice. This is research on past data, not advice to buy, sell or short anything. Short selling carries risk of losses larger than the amount invested.
Related Reading
- Do Top Gainer And Loser Lists Make Money?
- Do Classic Stock Signals Beat The Market?
- stkmrkt Short Portfolio
This is research on past data and not a recommendation to buy, sell or short any security. Past results do not predict future returns.
Frequently Asked Questions
Is shorting a stock after it drops 20% profitable?
Not reliably. Over 2013–2025, with realistic costs, the trade averaged -0.33% per trade with a range of -0.94% to +0.24%. It did beat shorting random stocks, but both lost money.
Why did it work in 2026?
2026 averaged +1.87% per trade across 540 trades, but the range included zero and one strong year is not evidence of a lasting edge. The 13 years before it were not positive.
What are the risks of shorting crashed stocks?
The loss is not capped, crashed stocks can bounce sharply, and many cannot be borrowed or carry very high borrow fees. Our test assumed every stock could be borrowed, which flatters the result.
Is this investment advice?
No. It is research on historical data and not a recommendation to buy, sell or short any security.