A 29.52% week at the 98th percentile tests a signal built on ordinary weeks
Should the research team credit this week as evidence the ETH signal works, suspend the signal, or keep it running at size while excluding the week from any re-estimation?
The market facts are real; the decision maker is illustrative.
The week and the position
In the week of 4 November 2024, ETH rose 29.52% from 2,457.73 to 3,183.21, a return at the 98th percentile of 351 weeks of history, with a 37.79% range and volume 2.51 times its prior 12-week average. The prior four weeks were mixed: +1.18%, +11.26%, -8.70%, -2.00%. ETH beat BTC by 12.66 percentage points, correlation with BTC was loose at 0.68, and funding stayed positive near 0.011% per eight hours. An illustrative team runs a directional ETH signal and must decide how to treat this week in its evidence base.
Judgment
A single week at the 98th percentile cannot validate or invalidate a signal; it is the tail the signal was never fitted to. Keep the signal running at its existing size, but exclude this week from any re-estimation of its parameters or of its win rate, and do not raise its size on the strength of the return. The team accepts that the week is uninformative and that its live results this week will be dominated by the outlier rather than by the signal.
What would change it
If the team's own logs show the signal was long ETH from Monday's low near 2,357.59 and held through Wednesday's 12.35% best day, this week counts as one genuine in-sample win, though still at reduced weight. If the next three weeks close with mixed signs near zero, the signal returns to normal treatment.
What sets this case apart
The decision is about whether an extreme week belongs in a signal's evidence base, not about forecasting whether ETH continues higher.