Deep-drawdown bounce (capitulation cohort, no MA dependency) · Null results

The result

Code: research/drawdown_bounce.py Report: research/drawdown_bounce_report.json (n=31,399 events)

Strips the MA cross out entirely: fires when a stock is 15 %+ below its trailing 252d high, has been falling for at least 60 trading days since that high, has a negative trailing-20d return, and is showing volume confirmation (5d vol ≥ 1.2x 60d baseline).

Bucket	n	60d edge	250d edge
dd_deep (-50% to -30%)	7477	+1.7 %	+6.4 %
dd_severe (≤ -50%)	1937	+7.6 %	+25.4 %
vol_confirmed + dd ≤ -30%	2654	+3.8 %	+12.5 %
OOS 2020+ vol_confirmed + dd ≤ -30%	1545	+4.3 %	+18.1 %
IS pre-2020 vol_confirmed + dd ≤ -30%	1109	+2.9 %	+3.8 %

Three observations: 1. The bounce is monotonically stronger with deeper drawdown. 2. The death-cross + dd combo earns ~3x the edge of dd alone — the cross does add timing on top of pure drawdown context. 3. n is much larger here than in the cross variant, so this is more practical to operationalise as a basket.

At a glance

Sample	31,399
Validator	research/drawdown_bounce.py
Verdict	Rejected

Every result here is reproducible from the script named above. Reports are in the repository.

Other rejected theses

All 13+ →