Inside Day Breakout — busted
A popular 79% inside-day-breakout claim from trading-education corners. We ran it on 5 years of NQ and found inside-day setups are *worse* than regular-day breaks of prior H/L. No model shipped.
updated 2026-04-21The claim
“Inside day” = prior day’s range is fully contained inside the day before it. Compressed range → pent-up volatility → breakout of the inside day’s high/low is supposed to print high close-direction continuation. A commonly cited figure on that claim: 79% close-direction success rate.
The test
audit_inside_day.py — n = 1,366 MNQ sessions. Baseline close-direction: 53.5% BULL / 46.5% BEAR.
| Cohort | n | Close-dir | Edge vs baseline |
|---|---|---|---|
| Inside-day-prev + break UP → BULL | 74 | 62.2% | +8.6pt |
| Inside-day-prev + break DOWN → BEAR | 68 | 54.4% | +7.9pt |
| Regular day + break UP → BULL | 609 | 64.0% | +10.5pt |
| Regular day + break DOWN → BEAR | 468 | 60.3% | +13.8pt |
| Inside day itself closes | 146 | 53.4 / 46.6 | 0 |
Inside-day cohorts are weaker than regular-day breakouts of prior H/L. The 79% figure does not hold in this dataset.
Compression buckets
We tried tightening the inside-day definition — narrower ranges, tighter compression ratios. Bucket sizes dropped to n=13–22 and results got noisier, not stronger. No compression band produced a meaningful edge lift.
Why the popular number is different
Best guess: the 79% claim comes from a small-sample or hand-curated set (charts that “look like” inside days in hindsight often include additional filters the publisher didn’t state). When you define the setup mechanically — prior day’s H and L both inside the day-before’s H and L — and run it on every instance over 5 years, the edge isn’t there.
What we shipped
Nothing. The generic PDH/PDL breakout signal — which is stronger on regular days than on inside days — is already covered by existing models. There’s no marginal alpha in adding the “inside-day” condition as a gate.
Lessons
- Pattern claims need mechanical definitions to be testable. “Inside day” is well-defined; “compression day” usually isn’t.
- Always compare against the generic version of the setup. Inside-day breakouts look impressive until you notice regular-day breakouts of PDH/PDL are better.
- Small cohorts amplify noise. Tight compression bands (n=13–22) will produce impressive-looking cherry-picks — don’t ship off them.
Related
- Herman 17-year Asia-London scan — larger-scale scan that did turn up real edges.
- Generic PDH/PDL break is the stronger cousin of this pattern — already part of the stack.