AI 中文总结
该研究分析七次BTC清算级联，发现临界预警信号具有事件异质性，吃单订单流方差压缩是总体前兆，单次事件临界减速论断脆弱，级联多为冲击驱动的不连续转变。
AI 中文摘要
加密货币永续合约崩盘是否携带有可复现的临界转变早期预警指纹，且该指纹存在于哪个状态变量中？我们利用分钟级价格数据和5分钟级杠杆率/订单流数据，对2022-2025年的七次重大BTC清算级联（包括2025年10月10日创纪录的190亿美元事件）展开研究。对去趋势残差，我们计算滚动方差和滞后1阶自相关，并采用Kendall-tau统计量检验其在级联前的趋势，对每个变量每次事件遍历39种分析配置。不存在对所有事件均不变的变量：价格在七次事件中的五次携带临界减速特征，但在恰好两次突发新闻（关税）冲击中无此特征，这表明存在两类结构：内生累积型级联与外生冲击型级联；2025年10月的事件，其样本内分析显示特征存在于杠杆率而非价格中，属于异常值而非普遍规律。所有有数据的事件中唯一留存的规律是吃单订单流方差的压缩，该压缩通过了300次起始安慰剂检验（费舍尔组合p值约为5e-6），但它是总体层面的前兆，而非单次事件的警报。因此，加密货币衍生品中单次事件的临界减速论断本质上是脆弱的。我们认为失败模式本身具有诊断性：在去稳定机制最突兀的地方，减速现象恰好不存在，这符合以下预期：这些级联是不连续的、冲击驱动的转变，而非临界转变。
英文摘要
Do crypto perpetual-futures crashes carry a reproducible early-warning fingerprint of a critical transition, and in which state variable? We study seven major BTC liquidation cascades (2022-2025, including the record 19B USD event of 10 October 2025) using minute-level price and 5-minute leverage/order-flow data. On detrended residuals we compute rolling variance and lag-1 autocorrelation and test their pre-cascade trend with the Kendall-tau statistic, sweeping 39 analysis configurations per variable per event. No variable is event-invariant. Price carries the critical-slowing-down signature in five of seven events but is silent in exactly the two sudden-news (tariff) shocks, suggesting a two-type structure: endogenous-buildup versus exogenous-shock cascades. The October 2025 event, whose in-sample analysis suggests the signature lives in leverage rather than price, turns out to be the outlier, not the rule. The one regularity surviving all events with data is a compression of taker order-flow variance, which passes a 300-onset placebo test (Fisher-combined p ~ 5e-6) but is a population-level precursor, not a per-event alarm. Single-event critical-slowing-down claims in crypto derivatives are therefore fragile by construction. We argue the pattern of failures is itself diagnostic: slowing down is absent exactly where the destabilising mechanism is most abrupt, as one would expect if these cascades are discontinuous, shock-driven transitions rather than critical ones.
Comments13 pages, 7 figures
