ZARQ · Model validation
Backtest Validation
▲
These are not live calls. ZARQ did not exist when these events occurred. This page reports how the crash model, run on historical data, would have flagged them — statistical validation of the method, never a claim that ZARQ called them at the time. The real, prior record is the Live Track Record.
Across 31,781 historical predictions, 7,456 exceeded the 0.5 crash-probability threshold; 29% of those went on to fall more than 50% — out-of-sample.
terra‑luna
model, 2022-02-28 · backtest
prob 0.78 → drawdown −100%
pepe
model, 2023-03-20 · backtest
prob 0.76 → drawdown −100%
Method
out-of-sample
7,456 high-confidence flags · 29% precision at >50% drawdown