A -15.85% week at the 4th percentile asks whether a signal failed or was tested outside its regime
Should the research team score this week as evidence that the signal failed, or reclassify it as an out-of-regime week and keep the signal at reduced size pending a test in ordinary conditions?
The market facts are real; the decision maker is illustrative.
The week and the position
In the week of 17 January 2022, BTC fell from 43,071.66 to 36,244.55 (-15.85%), a return at the 4.0 percentile of 205 weeks of history. It traded 27.93% from high (43,505.00, Thursday 16:00 UTC) to low (34,008.00, Saturday 10:00 UTC), closed in the lower third of that range, and did 13.69 billion of dollar volume, below its prior 12-week average. Friday alone fell 10.41%. A research team had a signal live and must decide what this week proves about it.
Judgment
A week in the bottom 4% of 205 weeks is not a fair sample for validating a signal built on prior weeks that averaged mixed returns and one up 8.77%. Volume below the 12-week average, and a largest single-hour fall of -4.36% that is smaller than Friday's -10.41% day, show the loss spread across the day's hours rather than arriving in one gap the signal could not have traded. Score the week separately, keep the signal running at a reduced size, and reserve judgment until ordinary weeks accumulate.
What would change it
If the team's own backtest shows the signal earns its edge specifically in weeks with realised volatility near 83.0% and returns in the bottom decile, this week belongs in the scored sample. If live trades this week lost beyond the backtest's worst drawdown, the signal is failed regardless of regime.
What sets this case apart
The question is whether an extreme week is evidence or noise, judged by the signal's own historical domain, not by the size of the loss.