Seasonal Pattern Detector (historical tendency)
What it shows
Recurring, statistically significant historical tendencies in prices: day-of-week seasonality (e.g. "BTC Mondays") and conditional probability (what historically happened AFTER a given condition). Each row describes the past with anti-noise safeguards: not a forecast.
How to read it
Read it like weather statistics, not a forecast for tomorrow. A +8 pp edge with n=120, a narrow CI and the "significant" plus "walk-forward" badges is a solid historical tendency. Treat a pattern without those badges or with a small n as a curiosity. This is NEVER a buy/sell signal, as past results do not guarantee future ones.
Parameters
- Base vs Conditional
- Base = the overall share of up days. Conditional = the share after the condition holds. The gap is the "edge" in percentage points.
- Positive reading: A clear positive edge (e.g. +8 pp or more) with the "significant" and "walk-forward" badges and n≥30: a historically strong up tendency (green arrow). This describes the past, not a forecast.
- Negative reading: A negative edge (red arrow): the condition historically preceded down days more often than average.
- Neutral reading: A small edge (a few pp) or no significance badges: treat it as a curiosity, not a tendency.
- n (sample size)
- How many historical cases back the pattern. We require at least 30: smaller samples are too noisy to trust.
- 95% CI
- The Wilson confidence interval for the conditional rate: a realistic range for the true value. Narrow = more confident.
- Seasonality charts
- The full "Seasonality" view shows four charts: average change by day of week (bars with whiskers), the average week path (a line accumulating from Monday), a month × year heat map, and volume by day of week (index 100 = the period average).
- Positive reading: A bar clearly away from zero with short whiskers (a narrow confidence interval) plus the "FDR-significant" and "Stable over time" badges: a tendency repeated across the whole window, not the effect of one year.
- Negative reading: Whiskers crossing zero (the confidence interval contains 0) or an "Unstable over time" badge: the chart shows noise, or a tendency that has already faded.
- Neutral reading: A faded bar = too small a sample for that group. Instead of a chart you may see "collecting data, we have X of Y days"; the history window is then still too short.
- Samples on a chart
- Above each chart there is "n = …": the number of days (for the heat map: complete months) the averages are computed from. More samples mean narrower whiskers and a more confident result; individual groups below the threshold are greyed out.
- Whiskers = confidence interval
- The vertical whisker on a bar (and the band around the week line) is the 95% confidence interval for the mean: the range that most likely contains the true value. If it spans zero, the difference may be chance.
- FDR-significant
- The pattern passed the Benjamini-Hochberg correction for multiple testing, which means it is not a lucky fluke among hundreds of tested conditions.
- Walk-forward
- The pattern held on data outside the window it was found in, which is evidence it is not overfit to history.
Example
"After a day dropping >5%, the next day was up 63% of the time (base 52%, n=88, CI 52–73%, significant, walk-forward)". Historically, bounces after sharp drops happened more often than average, but this only describes the past.