The Bid & Ask

Quotations, Dispatches, and the Daily Tape

Thursday, August 13, 2026

Rejected thesis

Bearish RSI / MACD divergence — first real SHORT alpha in the repo

The result

Code: research/bearish_divergence.py Report: research/bearish_divergence_report.json

Mirror of bullish_divergence: price prints a fresh higher pivot HIGH while the momentum oscillator (RSI or MACD histogram) makes a LOWER high. Buying momentum is exhausting even as price keeps grinding up.

Every other bearish chart pattern we tested turned out contrarian long (death cross, bear flag, downside gaps, bull pennant breakdown). Divergence is the exception — it's structural (a price-vs-momentum relationship), not a level reading, and it works symmetrically.

Bucket n 5d edge 20d edge down @ 20d
rsi_div all 18,040 -4.10 % -4.60 % 67 %
rsi_div runup > 50 % 9,310 -4.72 % -4.89 % 67 %
rsi_div runup > 100 % 3,216 -5.72 % -5.24 % 65 %
rsi_div OOS 2020+ runup > 100 % 2,022 -6.00 % -5.00 % 65 %
macd_div all 44,679 -4.15 % -4.46 % 67 %
macd_div runup > 100 % 5,184 -5.81 % -4.88 % 65 %

Why this matters: nearly every other "bearish" pattern in the repo is also a long signal (contrarian bounce family). Bearish divergence is the only structural signal that actually produces meaningfully negative forward returns. Magnitude is symmetric with the bullish version (~5 % edge), and stratifying by run-up (analogue of the drawdown stratification on the long side) sharpens it.

Practical use: pair with put spreads or short-stock positions. 67 % down-hit on n=18k is enough sample to productize as a bearish_div short cohort. Combine with VIX context (VIX < 20 + bearish_div = strongest case for the top).

At a glance
Sample18
Validatorresearch/bearish_divergence.py
VerdictRejected

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

Other rejected theses