The Bid & Ask

Quotations, Dispatches, and the Daily Tape

Thursday, August 13, 2026

Rejected thesis

Sector-relative drawdown overlay

The result

Code: research/sector_relative_dd.py Report: research/sector_relative_dd_report.json

Companion to the SPY-relative overlay. For each drawdown_bounce event, look up the stock's SPDR sector ETF and the ETF's own trailing-252d drawdown. The sector filter is sharper than the SPY one because SPY masks sector dispersion (a stock can be down 30 % while SPY is up 5 % but the sector is actually down 15 %).

Bucket × OOS 2020+ × vol_confirm n 60d edge 250d edge
sector AT HIGHS (sector_dd > -5 %) 333 -2.2 % +1.9 %
sector mild drawn ~ +3.2 % +11.7 %
sector deeply drawn (sector_dd ≤ -15 %) 649 +10.0 % +39.0 %

The "laggard catch-up" hypothesis (stock weak, sector strong → name rotates higher to match peers) is false — those stocks are genuinely idiosyncratic losers, not waiting-to-rotate laggards.

Live cohort recommendation: pair the SPY-dd gate with a sector-dd gate. Skip every drawdown signal when sector_dd > -5 %; demand sector_dd ≤ -10 % for highest-conviction entries. This filter likely removes most of the survivorship-bias risk from the deep-drawdown cohorts.


At a glance
Validatorresearch/sector_relative_dd.py
VerdictRejected

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

Other rejected theses