Bearish divergence is the warning reading: price sets a higher high while a momentum oscillator sets a lower one, and the claim is that buyers are having to push harder for each new high and are running out.
We can now test that claim mechanically. The backtester's divergence rule reads RSI(14) at confirmed 5-bar swing highs, and we ran it across 22 US stocks and sector ETFs the same way we test every signal in this series: buy the day the pattern becomes visible, hold ten sessions, and see how often that finished green. If the pattern has bearish content, buying into it should win less often than buying on a random day.
It printed 205 times; 13 arrived during open holds, and buying into the 192 the engine could take won 51.0%, against 55.4% for any random ten-day hold on the same names and dates. That is 4.4 points below the base rate: the direction the pattern's believers would predict, on a sample that cannot quite prove it.
What is bearish divergence
The two dashed lines are the whole signal: they point in opposite directions. Price is still setting new highs, but each one is being made with less momentum than the last.
- Price forms a higher swing high
- The oscillator, RSI here, forms a lower high at those same two swings
The reading is that the advance is thinning out: the second high, for all that it went further in price, drew less buying force than the first.
How the test avoids the usual cheat
A swing high is only a swing high once bars have printed on both sides of it. With 5-bar swings, the high of March 10th is confirmed on March 17th, and before that it is just an up day that might keep going. Backtests that date a divergence to the high itself are crediting the signal with a week of foresight nobody had. Our rule fires on the bar that confirms the second swing, the first day the pattern was visible on a chart, and every figure here is dated that way. The bullish article explains this trap in more detail; it applies identically in both directions.
What 4.4 points below the base rate does and does not mean
| Ten-day holds starting from | Windows | Finished green |
|---|---|---|
| any bar | 24,948 | 55.4% |
| a bearish RSI divergence | 192 | 51.0% |
Reading a gap like this needs care in both directions.
It is not proof. On 192 trades the 95% interval on that 4.4-point gap runs from about 12 points bearish to 3 points the wrong way (p is about 0.22). We ran the same buy-into-it test on One Black Crow, a bearish candle, and against this same 55.4% yardstick the crow came in 4.8 points below on 619 signals, which was distinguishable from chance (p is about 0.016). This is a similar-sized effect on a third of the data, so the same arithmetic that convicted the crow can only shrug here.
It is also not nothing. The gap points the way the pattern claims, the ten days after a bearish divergence were measurably not a good time to be buying in this window, and the bullish mirror came in above its baseline on the same names and dates. Two opposite patterns landing on opposite sides of the same base rate is the behaviour a real signal pair would show. It is weak evidence for, where most patterns in this series produced evidence of nothing.
What it clearly is not is a strong sell signal. A reader hoping divergence at the top calls corrections should notice the size of the effect even taken at face value: buying into the signal still won half the time, and the median name returned +2.9% doing it, at 7.0% exposure with a 12.3% worst drawdown. Buying into the bearish signal actually returned more than buying the bullish one (+2.0%), on a lower win rate, which is ten-day noise doing most of the work in both figures. The 192 trades also spread across 139 distinct days with no single cluster larger than six names, so unlike the bullish version, whose signals piled up 13 at a time at the 2022 bottom, tops arrived one stock at a time here. Divergence-at-the-top is a slow leak, not an alarm.
The oscillator matters
The same rule read on the MACD line instead of RSI took 186 trades and won 54.3%, a point below the base rate and indistinguishable from it (p is about 0.76). Whatever bearish lean the RSI version shows, the MACD version shows only a shadow of it in this window. If you read divergence off a specific indicator, test that indicator; averaging all oscillators into one concept averaged the signal away.
What we did not test
Every confirmed divergence counts, deliberately, because that is the pattern without a human filtering it. A chart trader would add context, and each piece is testable rather than trusted:
- At resistance that has rejected price before. A fading momentum high into a known ceiling is the textbook version.
- On the second or third divergence in a row. Practitioners argue repeated divergences compound; the rule as tested takes each on its own.
- As an exit rather than a short. Our test buys into the signal to measure it. Using it to take profits on a long you already hold is a different, gentler use, and the 51% figure is if anything more supportive of that: it says the easy gains were behind.
Test it yourself
The backtester carries this as a rule. In the free backtester, open Chart setup, add Divergence, and set it to Bearish. Point it at RSI, the MACD line or OBV, adjust the swing width and lookback, and run it on any ticker beside buy-and-hold. Inverting it into an exit rule for a long strategy is a five-minute experiment and a better use of the pattern than shorting it, on these numbers.
Scanning for them live
TC2000 can scan for RSI divergence across watchlists and alert when one forms. Finviz does not scan divergence directly, but its screener surfaces overbought stocks at new highs, which is where this pattern lives.
Neither will tell you what this page just measured, which is how much the warning was worth.
Frequently asked questions
Is bearish divergence a reliable sell signal? Weakly supported, unproven. Buying into 192 confirmed RSI divergences won 51.0% against a 55.4% base rate, so the ten days after a signal were somewhat worse than average, but on this sample the gap is not statistically distinguishable from chance.
Should I short a bearish divergence? Our numbers argue no. Even taking the 4.4-point gap at face value, buying into the signal still won half the time; a short needs much more than that to pay its costs. The gentler reading, taking profit on longs, is better matched to the effect size.
Does bearish divergence work better on RSI or MACD? RSI, on this test. The MACD-line version won 54.3%, a point below the base rate and statistically nothing, on the same names and window. One run, one window, but a reason to test your specific indicator rather than the concept.
How is this different from the bullish version? Mirror construction, different behaviour. Bullish divergence beat its baseline by 5.2 points and clustered hard at the 2022 market bottom; the bearish version undershot by 4.4 points and arrived scattered, one top at a time. Neither result reaches significance alone.
What is hidden bearish divergence? The continuation variant: a lower price high with a higher oscillator high, read as a downtrend resuming. We tested regular divergence, the reversal kind.
Before you trade this pattern
The figures here come from one run over 22 large-cap US stocks and sector ETFs, in a window that was mostly a bull market, which is worth weighing for a bearish signal: tops were rarer than bottoms in this sample. Signals are RSI(14) divergences on confirmed 5-bar swing highs, bought at the close of the confirming bar and held ten sessions; win rates are measured before costs, and return figures include commission and slippage. The engine holds one position per name at a time, so 13 of the 205 printed signals were skipped, and a hold still open at the end of the window is closed early. The script that produced them is in the repository (backend/scripts/research/candle_study.py) so the run can be repeated rather than trusted. Past prices are not a forecast, and nothing here is investment advice.