Lesson 17 of 20 · 6 min

Historical scenario replay, and its limits

Replaying real crises is a good discipline, and the sample is far too small to be a distribution.

Historical replay is a specific kind of stress test: instead of inventing a decline, apply the actual sequence of returns from a real episode to your current position.

Its advantages are genuine. Real episodes contain features nobody thinks to invent — the exact pace of the decline, the failed rally in the middle, correlations rising as it progressed, liquidity thinning at the worst moment. Those features are what determine whether a path touches a threshold, and a smoothly stipulated decline of the same total size may not touch it at all.

Its limits are equally real. The sample is tiny: a century of data contains perhaps six or seven genuine equity crises, which is not enough to estimate anything. Each episode had a specific cause that will not repeat in the same form. And the composition of a portfolio and the structure of the market have both changed, so applying 1973's returns to today's holdings is a rough analogy rather than a measurement.

There is also a subtle selection problem. The episodes in the record are the ones that happened, and the worst possible outcome has no reason to be among them. Treating the largest historical drawdown as a worst case assumes the past sampled the tail exhaustively, which nothing guarantees.

The right use of a replay is therefore as one scenario among several, valued for its realistic shape rather than as an estimate of frequency. What it answers well is: if something like this happened again, where would my threshold sit at each point along the way.

Key terms

Historical replay: Applying the actual return sequence of a past episode to a current position. Captures realistic pacing and path shape.
Small-sample limitation: The problem that only a handful of genuine crises exist in the record, so frequency cannot be estimated from them.
Worst case is unobserved: The point that the largest decline in the record is not the largest possible, and there is no reason to treat it as a bound.

Worked example

Peak-to-trough declines and the drawdown that triggers a call

Broad US equity drawdowns, approximate peak-to-trough. Trigger drawdown for the standard setup: 28.6%.

1973–1974: About −48%
2000–2002: About −49%
2007–2009: About −57%
2020, February to March: About −34%
Number of usable episodes in a century: Six or seven
What that sample can estimate: Shape, not frequency

Four of the last five decades contained a drawdown past this borrower's trigger. The sample is too small for a probability and large enough to be sobering.

Check yourself

What does a historical replay tell you that an invented 30% decline does not, and what can it not tell you?

Saved in this browser only. No account, nothing sent anywhere.

Educational only. Nothing in this lesson is investment, legal or tax advice, or a recommendation to enter into any transaction. Figures are illustrative and computed from the inputs stated in the lesson.