Liquidation cascades: why crypto moves further than the news justifies

Crypto perpetuals routinely produce moves whose size is unexplainable from whatever triggered them. The mechanism is mechanical rather than informational, and understanding it changes several design decisions in a bot.

The loop

1. Price moves against a cluster of leveraged positions.
2. Their maintenance margin is breached; the venue force-closes them.
3. Forced closes are market orders. They consume book depth and push price further.
4. The further move breaches the next cluster of positions.
5. Repeat, faster, until the leverage in that price region is exhausted.

Two properties make this violent. Liquidation prices cluster, because leverage settings cluster — a lot of people use 10× and 20× round numbers. And market makers widen or pull quotes when volatility spikes, so depth is thinnest exactly when the forced flow arrives.

What this does to a bot

Stops do not fill where you put them. During a cascade the price can print several percent beyond your stop level with no intermediate liquidity. Your loss is not the stop distance; it is wherever the market next traded.

Your own position can become the next domino. A position sized for normal conditions can be liquidated by a move that reverses within minutes — you took the full loss of a wick you would have survived at lower leverage.

Correlations go to one. Cascades propagate across instruments through shared collateral and cross-margin. Positions that looked diversified are, for those minutes, a single position.

Backtests do not contain them properly. Candle data hides the intra-candle path. A backtest assuming your stop filled at its level is describing a market that was not there.

Designing for it

The cooldown deserves emphasis. Strategies that re-enter straight after being stopped out during a cascade tend to enter into the reversal and get stopped again. A brief pause after any move exceeding N times normal volatility costs little and avoids the worst sequence.

Can you trade them?

Yes, and it is harder than it looks. The idea — provide liquidity into forced selling, capture the reversion — is sound and heavily contested.

The practical obstacles: you need depth data to know when the cascade is exhausted rather than pausing; you need to be flat before the next leg; and your own limit orders sitting in the path are what gets consumed if you are early. Sizing has to assume you are wrong about the bottom, because most of the time you will be.

Our honest view: this is a strategy for someone with real-time depth data, tested reversal logic and the discipline to take small size. As a first bot it is a way to lose money quickly.

The minimum defensive set

1. Leverage cap set by you, well below the venue's maximum.
2. Stop distance scaled by realised volatility, not a fixed percent.
3. A slippage budget on stops, so the loss estimate is honest.
4. Group exposure caps, because correlation is regime-dependent.
5. A cooldown after abnormal moves.
6. Liquidation distance monitored continuously and alerted on, per position.

None of these improve returns in normal conditions. All of them decide whether you are still trading after a bad ten minutes, which is a different and more important question.