Graveyard / Liquidations / Lighter
Liquidation cascade bots
Do liquidation-cascade trading bots work?
The idea
When open interest drops fast and price crashes, forced sellers are done. Buy the wick. The mirror: short after a short squeeze.
How we tested it
| Data | 30 Binance USDT perpetuals, 5-minute open interest and prices. |
|---|---|
| Period | Data from April 2023 to October 2026. Test window October 2024 to October 2026, out-of-sample from 16 May 2026, plus an older holdout from June 2023 to October 2024. |
| Costs | 0.13% per round trip. Also re-run at 1.5 times that. |
| Holdout | The older 2023-24 holdout and the out-of-sample stretch since May 2026. |
| Control | No random-entry control. Instead the three best days were removed to see what is left. |
Result
| Trades per week | 15.0 |
|---|---|
| Net per trade | +19.9 bp |
| Day-clustered t-stat | 0.86 |
| Without the 3 best days | −89% |
| Short-squeeze mirror | −63.7 bp per trade (n 231) |
Why it failed
Almost all the profit came from one crash day. Adding coins only added more exposure to that same afternoon. At aggressive sizing the chance of a 50% drawdown was 93%. Shorting squeezes lost in both halves and in the holdout.
What would make it worth retesting
No for the price-and-OI version. A trigger based on the size of liquidations themselves is still untested, because that data only starts in October 2026.
Test id: liqfam, liqfam#2, liqfam#3. Past results of a backtest say nothing certain about the future.