⚠️ Personal research and trading journal — not investment advice. The author does not provide licensed advisory services.
Finding a real statistical signal is rare. Turning it into a tradeable edge is harder than it sounds. This article is about a signal that is demonstrably real — where the confidence interval excludes zero, where a dose-response relationship exists, where removing outliers doesn't destroy the effect — but which fails as a standalone trade after realistic costs.
The lesson is about the gap between "statistically significant" and "tradeable." These are not the same thing.
The Signal: Climax Bearish Divergence
Climax conditions in a stock happen when price behavior becomes extreme: - Price accelerates above the prior channel (gap up, largest daily gain in weeks) - Volume spikes to multi-month highs - The stock extends far above all moving averages (often 30%+ above 200d)
In a climax, everyone who wants to own the stock already does. The remaining buyers are late momentum chasers. The stock is pricing in perfection.
Bearish divergence adds a technical component: price makes a higher high, but an oscillator (relative strength, momentum, or volume-weighted metrics) makes a lower high at the same time. The price move isn't being confirmed by the underlying momentum.
The signal I tested: climax conditions plus bearish divergence at the climax peak. Both conditions simultaneously. Climax alone is not the signal. Divergence alone is not the signal. The combination is the signal.
Why I Expected This to Be Noise
Most short signals on individual stocks fail in backtesting. The combination of borrow costs, timing difficulty, and the tendency for climax stocks to continue higher before reversing makes short-selling these setups extremely hard to execute profitably.
I expected to run the numbers, find a weak or negative statistical signal, and file it as falsified.
That's not what happened.
What the Statistics Show
Testing the climax bearish divergence signal on US stocks 2010-2025:
Forward returns after signal (long short of the signal stock):
| Horizon | Median return | CI (95%) | % Positive |
|---|---|---|---|
| 5 days | −0.78% | [−1.12%, −0.44%] | 41% |
| 10 days | −1.45% | [−1.98%, −0.91%] | 39% |
| 20 days | −2.10% | [−2.78%, −1.35%] | 36% |
The forward returns are negative. The confidence intervals exclude zero at all horizons. A drop-top-3 robustness check (removing the three largest outlier moves) doesn't change the sign or significance. A dose-response test — sorting by "climax intensity" — shows monotonically increasing negative returns as the climax is more extreme.
By statistical standards, this is a real signal. Price reliably underperforms after climax bearish divergence.
Why the Short Still Fails
Running this as a naked short strategy — short the stock after the signal, hold for 10-20 days, cover — produces:
- Kelly fraction: −0.71 (the Kelly criterion says DON'T trade this)
- Sharpe ratio net of costs: −0.23
- Win rate: 36% on individual trades
The issue is the distribution. Stocks in climax conditions that then reverse do indeed come down. But a meaningful subset don't reverse immediately — they have one more leg up, sometimes 15-25%, before the eventual decline. That subset kills the short position before the reversal arrives.
The median is negative, but the mean is much less negative (or even slightly positive) because a few large against-you moves dominate the short side. A short position can't survive the whipsaw.
After transaction costs and borrow fees, the strategy generates a negative Sharpe. It's not tradeable as a standalone short.
The Correct Use: Exit Signal for Longs
The signal only becomes useful when framed from the long side.
If you're holding a stock that has made a significant move for you — say, 25-50% in your favor — and you see climax conditions combined with bearish divergence at a new high, this is a signal to reduce or exit your long position.
You don't need to short the stock. You just need to stop being long it. The expected forward returns over the next 10-20 days are negative, which means your open gain is in danger of reverting.
The implementation: 1. Flag a climax condition: the stock's daily gain is the largest in 20 days, volume is 2× the 50-day average, and the stock is 20%+ extended above its 200d MA. 2. Check for divergence: the 14-day RSI is lower at this new price high than it was at the prior high (bearish divergence). 3. If both conditions met: this is a signal to take partial or full profits on your long position.
This is how the signal is wired into the system: as a flag on the current position tracker, not as a short entry.
The Broader Principle
CI excluding zero is a necessary condition for a signal to be real. It's not a sufficient condition for a signal to be tradeable.
To be tradeable, a signal also needs: - A distribution that survives costs: the mean and median both need to be sufficiently negative (or positive) that transaction costs don't erode the edge. - Timing that matches execution: a signal that is statistically accurate but requires entering at a price you can't actually get (market impact, gap opens) fails in practice. - A mechanism you can use: "short a climax stock" requires borrow availability, margin, and tolerance for against-you moves that many traders (including me) don't have.
The climax bearish divergence signal passes the first test (CI excludes zero) but fails the second and third for short-side trading. What makes it valuable is finding the use case where those constraints don't apply — exit for longs, where you own the stock already and can simply sell.
Finding the right use case for a signal is as important as finding the signal itself.
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. Signal: climax conditions (largest daily gain in 20 days + volume ≥2× 50d avg + price ≥20% above 200d) PLUS bearish divergence at new high (14d RSI lower than prior high). Tested US stocks 2010-2025. CI from bootstrap resampling. Costs include commission + borrow. Kelly and Sharpe calculated on realistic execution.