Sornette

We built fatcrash, a Rust+Python toolkit with 15 crash detection methods: LPPLS, DFA, EVT, Hill, Kappa, Hurst, GSADF, momentum/reversal, price velocity, and more. Tested on 96 drawdowns across BTC, SPY, Gold, 23 forex pairs, and equity crises with honest precision/recall/F1 metrics. Plus: which methods transfer to revenue and profit data.

Everyone agrees markets react to themselves. The question is how much. A statistical tool built for earthquakes turns that vague idea into a single number: the fraction of market activity that is the market reacting to its own moves rather than to outside news. The number turns out to be close to the level where a chain reaction would run away.

The last essay said big cascades are built in and the trigger tells you nothing. Didier Sornette disagrees, at least about the biggest ones. He argues a bubble grows faster than exponentially toward a finite-time singularity, leaves a telltale wobble on the way up, and that this makes some crashes partly foreseeable. This is the optimistic case, and its limits.