Divergence is when price and an indicator disagree: price makes a higher high while RSI makes a lower high (bearish), or price makes a lower low while RSI makes a higher low (bullish).
It is the most sophisticated-sounding tool in retail technical analysis. Here is what it measured.
We defined it mechanically over a 20-session lookback: price at least 5% higher (or lower) than 20 sessions ago, while RSI moved at least 5 points the other way.
| sample | touched +20% | vs the market | |
|---|---|---|---|
| Bullish divergence, TRAIN | 879 | 20.7% | 1.18× |
| Bullish divergence, TEST | 535 | 47.9% | 1.66× |
| Bearish divergence, TRAIN | 1,141 | 25.9% | 1.47× |
| Bearish divergence, TEST | 1,360 | 30.4% | 1.06× |
Two very different stories.
Bearish divergence: decayed to nothing
1.47× in the design years, 1.06× in the held-out years. That is the signature of a pattern that was an artefact of one era. Treat it as unproven. Note also the direction problem, "bearish" divergence was associated with higher odds of a +20% rise in both periods, which should tell you the label itself is not measuring what its name claims.
Bullish divergence: promising, and unstable
1.18× then 1.66×. Positive in both periods, which is more than most rules in this course manage, but the size more than doubled between them. And in the held-out years the median 20-session return for these setups was +7.56%, the highest figure anywhere in this study.
The strongest version we found was bullish divergence on a stock below its 200-day average: 1.71× in the held-out years with a +8.50% median 20-session return, on 488 observations. But the same setup measured only 1.16× in the design years.
How to hold a number like this honestly: a rule that measures 1.16× then 1.71× has not been validated. It has been observed twice with different strengths. It is a candidate, not a conclusion. ZEME does not put candidates into the engine, and we are not asking you to either. We are showing you the working.
That is the difference between this course and every other one: the 1.71× is exactly the number a marketer would put on a banner. We are putting it here, next to the 1.16× that undermines it.
Do this now, practical activity
- Open SuperCharts with RSI enabled on a stock that has fallen a lot this year.
- Find a point where price made a lower low than a month earlier, but RSI's low was higher than its earlier low. That is a bullish divergence.
- Mark what happened next.
- Now count how many times you had to look before you found one. That number matters as much as the outcome. A pattern you have to hunt for is a pattern you will also imagine.
The trap this activity is built to expose: divergence is easy to see in hindsight and hard to see in advance, because you get to choose which two highs to connect. Any rule where you choose the inputs after seeing the outcome will always look excellent on a chart and fail in live trading.
Measured: bullish divergence touched +20% in 20.7% of 879 cases (1.18×) and 47.9% of 535 held-out cases (1.66×), positive in both periods but unstable in size. Bearish divergence decayed from 1.47× to 1.06× and is treated as unproven.