Research · test #2
Do crypto pumps always dump? No edge
"Every pump gets dumped" is one of the most repeated lines in crypto. We took every pump on Binance futures over six months, wrote the rules down before loading any data, and checked. Pumps did not dump more often than chance, and shorting them made nothing on average - while a few trades lost more than the whole position.
In short: 1,215 pumps of +20% or more on 397 coins (21 Apr - 20 Oct 2026). Price reached a 50% pullback before an equally distant level higher in 52.0% of cases (random: up to 53.4%). A short at the signal won 69% of the time with a median of +8.8%, but the average after costs was only +1.1% (t = 1.03) because 27 trades lost more than 100%.
The rules (fixed before the run)
- Coins: all 522 crypto perpetuals on Binance USDT-M futures (tokenized stocks and commodities excluded), hourly candles.
- Pump: the hourly close is at least 20% above the lowest close of the previous 12 hours, on 12-hour volume at least 3 times the 7-day norm. One event per coin per 72 hours. Dump: the mirror image.
- Entry price is the close of the hour when the pump becomes visible - not the top, which nobody knows in real time.
- Test: within 72 hours, does price first reach a pullback of half the move, or a level the same distance further up? If pumps really "always dump", the pullback should win clearly more than half the time.
- Edge requires: at least 55%, p < 0.01, above what random side assignment produces, on both the first four months and the last two, plus a positive trade after costs.
Result: no edge, either way
| Events | Pullback first | p-value | Random, 99th pct. | |
|---|---|---|---|---|
| Pumps, Apr-Aug | 791 | 50.8% | 0.335 | 53.9% |
| Pumps, Aug-Oct | 398 | 54.3% | 0.049 | 55.5% |
| Pumps, all | 1,189 | 52.0% | 0.091 | 53.4% |
| Dumps, all | 404 | 54.5% | 0.041 | 55.4% |
Events where price hit neither level within 72 hours, or both inside one hour, are left out of the share (26 pumps, 120 dumps). Splitting by whether Bitcoin was rising or falling at the time changes nothing (51.5% vs 52.8% for pumps).
Why "90% of pumps retrace" sounds convincing
Because it is almost true - and almost meaningless. 76.9% of our pumps gave back half the move within 72 hours. But 59.5% also went the same distance higher within the same 72 hours. Big moves come with big swings in both directions; counting only the pullbacks measures volatility, not a pump-and-dump pattern. After the signal hour the median pump still ran another +11.2% within a day, and more than half ran at least 10% further.
Shorting the pump: wins often, loses big
A short opened at the close of the signal hour and closed 72 hours later, with 0.15% for fees and slippage:
| Trades in profit | 69% |
| Median result | +8.8% |
| Average win / average loss | +16.1% / -32.7% |
| Lost more than 50% of the position | 72 trades (5.9%) |
| Lost more than 100% (price more than doubled) | 27 trades (2.2%), worst -414% |
| Average after costs | +1.1% (t = 1.03 - not distinguishable from zero) |
Most of the time the short looks like easy money. Then a coin doubles or quadruples in three days and takes back everything the other trades made: without the worst 1% of trades the average would be +3.4%, without the worst 5% it would be +7.3%. With leverage, every one of those tail trades is a liquidation (see how little it takes in the liquidation price calculator). And this is before funding, which during pumps is often strongly negative - that is, paid by shorts.
And buying the dump?
The mirror image, with the same verdict. After 524 drops of 20% or more, buying at the signal and holding 72 hours lost money in 59% of trades (median -4.2%); the average of +1.3% comes from a few huge rebounds and is not significant (t = 0.71).
What this test does not say
- Coins delisted before 21 October 2026 are missing. Those are exactly the coins whose pumps most likely collapsed, so the real result for shorting pumps may be somewhat better than shown. We cannot say by how much.
- Hourly candles: a faster entry, a stop-loss or a different holding time is a different strategy, not tested here.
- A price spike with heavy volume is not proof of a coordinated pump-and-dump scheme. Organized pump-and-dump schemes are market manipulation and illegal in many countries; this test only looks at what price did after the spike.
Data and method
Rules were committed before any data was loaded; the full rules, code and every event are kept with the project, and we will share the event list on request. Data: Binance public futures candles. The same method as our first test: a control level at the same distance, shuffled sides, an in-sample and an out-of-sample part. Live scanner with the same rules: crypto pump & dump scanner. Live data for any coin: liquidations, open interest, funding rates.
FAQ
Do crypto pumps always dump?
No more often than a random price path would. Price reached a 50% pullback first in 52.0% of 1,189 resolved pumps - inside the range random chance produces.
Is shorting a crypto pump profitable?
It wins often and loses big: 69% of shorts were profitable, but 2.2% lost more than the whole position, and the average after costs (+1.1%) is statistically indistinguishable from zero - before funding.
Why do people say 90% of pumps retrace?
Most big moves retrace partly regardless: 76.9% of pumps gave back half the move in 72 hours, but 59.5% also went the same distance higher. The retrace figure alone measures volatility.
Do coins bounce after a crash?
Not reliably: the bounce came first in 54.5% of resolved cases (not significant), and buying the dip lost money in 59% of trades over 72 hours.