Relative drawdown vs SPY — critical macro overlay

Code: research/relative_drawdown.py Report: research/relative_drawdown_report.json

Splits the drawdown_bounce cohort (n=9,414 events with stock dd ≤ -30 %) by SPY's drawdown at the same date:

Idiosyncratic deep drawdowns have essentially zero edge. The drawdown bounce thesis is really a macro mean-reversion play disguised as a single-name pattern — when SPY is at highs and a single stock is down 30 %+, the market is usually right and the stock keeps falling (value trap, fraud, dying franchise). When SPY itself is down, the broad recovery sweeps the over-sold names back up.

Live cohort implication: every member of the contrarian family (drawdown_bounce, gap_capitulation, bounce, bullish_engulf_in_drawdown, consensus_capitulation) should add a SPY-drawdown gate — only fire when SPY is at least mildly drawn down (e.g. SPY dd ≤ -5 %). That single filter likely 2-3× the per-trade edge and removes the value-trap left tail.