Gap-pattern survey (breakaway / runaway / exhaustion / island)

Thesis (mixed): Edwards & Magee classified opening gaps by where they sit in the trend — breakaway (after consolidation), runaway (mid-trend continuation), exhaustion (end of trend, reversal), island (gap + opposite gap = reversal). Each is supposed to imply a specific forward direction.

Code: research/gap_patterns.py, research/gap_island_combo.py Reports: research/gap_patterns_report.json, research/gap_island_combo_report.json

Headline (unconditional): All eight gap variants show small edges in the 0.1 - 3 % range over 5-40d. The bullish gaps (breakaway_up, runaway_up, exhaustion_up, island_down_reversal) are roughly flat; the bearish gaps systematically have positive 20-40d edges — i.e., textbook bearish gaps are contrarian long signals at the daily resolution, same family as bear flag / death cross.

Why most fail: At the unconditional level the gap by itself isn't big enough to time the reversal — only ~1-3 % edge. After options friction these compress to ~0.

Conditioning on drawdown ≤ -30 % + OOS 2020+ (gap_island_combo.py) amplifies every bearish-gap variant substantially:

Verdict on the gap family: - Unconditional: skip. - Bearish gaps conditioned on dd ≤ -30 %: real edge, same capitulation-bounce family as death-cross / drawdown-bounce. Worth productizing as additional timing triggers for the broader drawdown-bounce cohort (multiple triggers → higher conviction when several fire near simultaneously).