Stress test plans against outcomes beyond the historical range

Ask how your plan holds up if the worst outcome is twice as bad as any historically observed case.

Why it works

Ludic fallacy risk models use historical distributions to set worst-case scenarios — but in real-world domains, the historical maximum is not the maximum. Because models are fitted to observed data, they will always underestimate the tail if the tail is determined by processes that were not active during the observation period. Stress testing beyond the historical range forces the plan to confront outcomes that a model cannot generate by construction.

How to do it

1. Take your worst-case scenario from your risk model.
2. Double it or triple it: "What if the losses are 2–3x our worst observed case?"
3. Check whether the plan survives at that scale, and build the plan toward surviving it if possible.

Evidence

Post-crisis analysis consistently finds that pre-crisis risk models underestimated tail outcomes, often because the models were built from data that did not include the crisis regime. The 2008 financial crisis is Taleb’s canonical example. (observational)

Stress testing beyond historical range is now standard in financial regulation (regulatory stress tests) but adoption in everyday planning is limited. The practice is principled but the specific multiplier (2x, 3x) is a heuristic.

Common mistake

Choosing a stress test level that is slightly worse than the historical worst case rather than genuinely outside it — this exercises the model’s known territory, not the territory beyond it.