▲ Work in progress. This site is built and researched in the open: pages, numbers and findings may change or be wrong, and the record has gaps. Not advice. More
The Lab
Ideas traders use, put to the record
← the Lab · signals tested · write-up
#36 Buying a sharp drop: does the market bounce the next day? Failed on fresh data
Past prices; no real orders were placed. Every figure is a share of the money, or a count.
Last checked: 16 October 2026.
Where the idea comes from
- A sharp move followed by a partial bounce is "short-term reversal", a form of mean reversion. Narasimhan Jegadeesh and Bruce Lehmann documented it in stocks in 1990.
- Its history, the maths step by step and its catch are on our theory page "Mean reversion: does what drifts away come back?".
- Testing a lead again on prices never used before is the standard guard against luck. The lead came from our study "Forced selling and crowded trades: open interest, funding, long/short ratios".
The question
Our study of forced selling left one lead: in 2022–2026, after a sharp one-hour drop, the market rose more than usual over the next day. Short bounces like this ("short-term reversal") were documented in stocks, by Jegadeesh and by Lehmann in 1990. But that lead came from a check added after we had seen the results, and that study's test year was already used. A lead found that way can be luck. So we tested it again on prices none of our studies had touched, 2020 and 2021, with every rule written down before those prices were read.
How we tested
- Two markets, each judged on its own:
- Bitcoin;
- an equal-weight basket of the 26 coins (of the 54 in the first study) that had futures then.
- The drop: an hour worse than the first study's worst 1% of hours, −1.70% for Bitcoin and −2.54% for the basket. These came from 2022–2025 and were fixed before any 2020–21 price was read.
- The trade:
- buy at the end of the drop hour, sell 24 hours later;
- pay the market-order fee both ways (0.055% each), and the funding;
- at most one trade a day, never two at once.
- Two comparisons, 2,000 draws each:
- same day: a random hour of the same day instead;
- same hour: the same hour of the day, on a random day within a month either side.
- Both share the market's trend, which in 2020–21 was strongly up. To count, the real trades had to beat both.
- To pass: at least 20 trades, a profit after costs, and beating 95% of both comparisons' draws, with and without March 2020 (the COVID crash).
The maths, step by step
1. A drop. Bitcoin at 50,000 falls to 49,150 within an hour: −1.7%, past the mark.
2. One trade. Buy at 49,150 and sell the next day at 49,500: +0.71%. Take off the fee both ways (0.11%) and three funding payments of 0.01% (0.03%): +0.57%.
3. Against the same hour on other days. Over the 192 real trades, Bitcoin made +0.26% a trade. Buying at the same hours on nearby days made +0.12%. The difference, +0.14%, is what the drop itself added.
4. Could that be luck? Of 2,000 sets of "same hour, nearby day" trades, the real trades did better than 65%. So luck alone does as well about a third of the time. The bar was 95%.
What we found
1. No bounce beyond the market's rise. Neither market passed.
| trades | real, a trade | same hour, nearby days | beaten | |
|---|---|---|---|---|
| Bitcoin | 192 | +0.26% | +0.12% | 65% |
| Bitcoin without March 2020 | 177 | +0.33% | +0.21% | 64% |
| Basket | 170 | +0.73% | +0.34% | 83% |
| Basket without March 2020 | 158 | +0.91% | +0.44% | 87% |
- About half of each profit was 2020–21's rise: buying at that hour on any nearby day earned it too.
- What the drop added was never clearly more than luck.
2. One of our two comparisons was the wrong one.
- The same-day control lost about 1% a trade, so the real trades beat it every time. That means nothing. A crash at 15:00 is a bad day: a random hour earlier that day, say 09:00, has the crash inside its next 24 hours.
- So the same-hour comparison alone decides the result, and it fails. We set both in advance and report both.
3. The risk was large. The worst single trade lost 47%, in March 2020, on Bitcoin and on the basket. Without that month, the worst was −21% (Bitcoin) and −31% (the basket).
What it means
- The bounce seen in 2022–2026 didn't hold up on fresh prices. Buying sharp drops earned about what buying at that hour earned anyway, with crash-sized losses along the way.
- This is the usual fate of a pattern found after looking at the results, and why we test such leads again on data kept apart.
Limits
- The 2020–21 basket had 26 coins, the 2022–25 basket 54. A smaller basket swings more, so it reached its drop mark a little more easily.
- The basket is put back to equal weights every hour, which no real book does for free.
Update, 16 October 2026: a correction
Added after this page was published. The text above is left as it was.
- "About half of each profit" held for three rows of four. The same-hour trades on nearby days earned 46% to 48% of the real trades' profit for Bitcoin, the basket, and the basket without March 2020. For Bitcoin without March 2020 it was 64%, nearly two thirds.
Credits
- Narasimhan Jegadeesh, "Evidence of Predictable Behavior of Security Returns", Journal of Finance, 1990; Bruce N. Lehmann, "Fads, Martingales, and Market Efficiency", Quarterly Journal of Economics, 1990: short-term reversal.
- Nicholas Metropolis and Stanisław Ulam, "The Monte Carlo Method", Journal of the American Statistical Association, 1949: drawing random comparisons many times.
- The lead itself came from our study "Forced selling and crowded trades: open interest, funding, long/short ratios". The re-test design is our own.
- Data: Binance's public archive of 1-hour USDT perpetual prices and funding.
- PerpGremlin and Claude, Buying a sharp drop: does the market bounce the next day?, TerminalGremlin, 2026: the tests, the code and the words.
← Forced selling and crowded tradesElliott waves →
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.