A new analysis of seven major liquidation cascades between 2022 and 2025 is casting doubt on the reliability of crypto crash prediction models that rely on gradual pre-crash "critical slowing down" signals. The study, which examined data from events including the October 10–11, 2025 crash that saw $19.1 billion in forced liquidations in a single day, found that cascade onsets were abrupt rather than gradual, and warning signals were inconsistent across the seven events. The researchers propose a two-type classification: endogenous build-up cascades and exogenous shock cascades, with different early-warning characteristics.
What the data shows
Price-based critical slowing down appeared in five of the seven cascades but was absent in the two sudden-news tariff shocks. In-cascade analysis revealed that the order parameter jumped by between 1.6 and 4.4 baseline standard deviations at onset across all events. A susceptibility proxy collapsed in five of seven events, and no event showed diverging susceptibility — a key signature that critical slowing down models typically expect. According to CoinGlass's 2025 derivatives market annual report, total forced liquidations for the year reached $154.6 billion, with a single-day peak of $19.1 billion during October 10–11, affecting about 1.6 million traders.
The October 10–11 crash in detail
During that October crash, futures led the move. BTC futures basis swung roughly $1,367 in eight minutes, trading volume spiked about 22 times baseline seven minutes before the trough, and the mark price undershot both spot and futures, feeding a reflexive liquidation loop. The study notes that 88% of all post-onset forced selling landed within 30 minutes, and 63% of that selling was absorbed off-book by the venue's backstop. Open interest cleared by 25 to 70% during cascades. A forward-looking microstructure metric called Slippage-at-Risk (SaR), calibrated on Hyperliquid order-book data and tested on the October 10 event, demonstrated leading-indicator properties for systemic stress.
Two types of cascades
The findings suggest that early-warning signals built on critical slowing down are regime-dependent. In endogenous build-ups, pre-state deterioration in price statistics may help. But in exogenous shocks, pre-state signals are unreliable, and forward-looking liquidity risk measures become central. The study emphasizes that microstructure and exchange design are first-order drivers of realized outcomes, not just investor positioning or macro news flow. This challenges the common assumption that crash prediction is primarily about market sentiment or external news.
What risk teams should watch
Researchers recommend that risk teams treat regime identification as a first step. Indicators to prioritize include order-book depth-at-risk, projected slippage for forced flow, basis behavior between spot and perps, and sensitivity of the mark price to thin prints. The SaR metric, which quantifies slippage risk under forced flow, could serve as a real-time complement to traditional early-warning signals. The study's authors argue that relying solely on price-based critical slowing down signals could leave risk teams blind to exogenous shocks. The next step, they say, is to integrate regime classification into real-time monitoring systems.
