One 19.23% day inside a 97th-percentile week asks whether a signal was tested at all
Should the research team mark its ETH signal as validated by this week, or record it as a single-week outlier and keep the signal unconfirmed?
The market facts are real; the decision maker is illustrative.
The week and the position
In the week of 20 May 2024, ETH rose 24.59%, a 97th-percentile weekly return against 327 weeks of history. The best day, Monday 20 May, rose 19.23%, and the largest one-hour move was 9.21% late that same UTC day. The worst day lost only 1.46%, and ETH beat BTC by 21.22 percentage points with an ETH/BTC ratio change of 20.53% at loose 0.6 hourly correlation. Volume ran 1.49 times the prior 12-week average, funding averaged 0.0102% per eight hours, and the four prior weeks were mixed (two up, two down).
Judgment
A week this extreme is a stress observation, not a clean test. The best day rose 19.23% and the largest one-hour move, 9.21%, landed late in that same UTC day, while the worst day lost only 1.46%, so one session dominates the weekly record. Record the week, but do not raise confidence in the signal or its position size on a single 97th-percentile observation. Separate the two kinds of evidence: ETH finished 21.22 percentage points above BTC at loose 0.6 correlation, a different finding from the absolute 24.59% gain, and the two should not be credited as one test.
What would change it
If the signal's backtest shows similar single-session concentration in most of its winning weeks, this concentration is not disqualifying. If the signal would not have been positioned before Monday's 19.23% day, this week's evidence for it should be discarded entirely.
What sets this case apart
Validation is judged on how the week's return is composed, not on its size: one best day inside a 97th-percentile week can confirm nothing.