⚠️ Personal research and trading journal — not investment advice. The author does not provide licensed advisory services.
In May 2026 I completed a test of climax bearish divergence — a pattern where a stock makes a new price high but its momentum indicator (RSI) fails to confirm, making a lower high. The hypothesis was that this divergence predicts near-term weakness.
The statistical signal was real. The trade I thought it implied — a short — was not.
What the Data Showed
Across the test, climax bearish divergence produced a statistically significant downward signal in 30-day forward returns:
- CI excludes zero: the confidence interval for the median delta was [-0.78pp, -2.10pp], well below zero
- Dose-response monotonic: the effect strengthened with divergence severity (larger RSI gap = larger forward return delta)
- Survived robustness test: when I dropped the top 3 outlier observations, the signal remained statistically significant
In total the signal cleared what I call Principle #8a: CI excludes zero is necessary but not sufficient for a tradeable signal.
The reason for that caveat: the Kelly criterion calculation came out negative (-0.71) and the Sharpe ratio after net transaction costs was -0.23. A statistically real signal with negative Kelly is a signal you cannot bet on profitably at any position size.
Why? The signal's downward prediction is real but the distribution of outcomes has too many cases where the stock continued upward after the divergence, often sharply. The mean effect is negative; the variance is too high to trade the short side.
The Correct Interpretation
Climax bearish divergence is real. As a naked short signal it fails. As an EXIT signal for existing long positions, it earns a different evaluation.
If I hold a stock that: 1. Has made a new price high 2. Has RSI failing to confirm (lower RSI high) 3. Is at or near a multiple of my initial risk
...that is a signal to tighten my exit, not to initiate a short.
The mechanism that makes the short unprofitable (too many upward continuation cases) makes the exit signal useful. I don't need the stock to go down for the exit to be correct — I need the exit to protect gains. If the stock continues up after I exit, that's fine. If it reverses, I avoided the reversal with evidence behind the timing.
This is now wired as a soft-tag in my position management: climax divergence conditions trigger a "prepare to exit at next weakness" note, not an automatic sell. Context matters — if the stock is at 1R profit, divergence doesn't override the trade thesis. At 3R with divergence present, that's near the exit window anyway.
Principle #8a
This finding generalized into a principle I now apply to all signal testing:
CI excludes zero ≠ tradeable.
A signal can be statistically real and economically unviable simultaneously. The conditions where this happens: - High variance in the outcome distribution (many outliers in the wrong direction) - Net costs that eat the small average edge - Directionality is correct on average but the loss cases are larger than the win cases
When I test a signal and CI excludes zero, the next question is: what does Kelly say? And what does Sharpe say after realistic transaction costs? Those three together determine whether the signal is real, interesting, and usable.
The climax divergence is real and interesting. As a short, it's unusable. As an exit prompt for longs, it's valuable.
Track. Study. Wait. Strike.
Personal research and trading journal — not investment advice. The author does not provide licensed advisory services. — MOEasymmetry
Draft 2026-06-12. Source: Climax bearish divergence test completed 2026-05-17. CI [-0.78pp, -2.10pp], monotonic dose-response, survived drop-top-3. Kelly: -0.71. Sharpe (net costs): -0.23. Conclusion: naked short FAILED. Use as long exit signal only. Full methodology at vault concepts/Climax-Bearish-Divergence-2026-05-17.md. Principle #8a formalized from this test.