Weekend effect
What it shows
A comparison of the weekend (Saturday+Sunday) with weekdays: mean change, 95% confidence interval, share of up days and badges for both groups, computed from the same daily series as the other models.
How to read it
What matters is the DIFFERENCE between the groups, not the values themselves: if the two confidence intervals overlap heavily, the weekend did not differ meaningfully from weekdays historically. Crypto trades 24/7, so the “weekend” is a calendar convention, not a trading break.
Parameters
- n (sample size)
- How many historical days back the group. A small n (the “low sample” badge) = a result too noisy to take seriously.
- Whiskers / 95% CI
- Whiskers are the 95% confidence interval of the mean: a realistic range for the true value. A narrow interval = a more confident reading.
- Positive reading: The whole interval on one side of zero: a clear tendency across the data window.
- Negative reading: Whiskers crossing zero: the interval contains 0, so it may just as well be noise.
- FDR-significant
- The green badge: the result passed the multiple-comparison correction (Benjamini–Hochberg). Without it, with ~30 tests per profile, a few “significant” results are pure chance.
- Stable over time (walk-forward)
- The stability badge: the effect keeps the same direction in both the earlier and the later half of the window. “Unstable” = the tendency may have already faded.