RSI Divergence vs RSI Overbought: 92 Signals Beat 436

Almost everyone meets RSI the same way: a line that goes above 70 and below 30, with the promise that those levels mark exhaustion. Divergence is what people graduate to when the levels disappoint them. That makes for an obvious question with a measurable answer — on identical data and identical rules, is the graduate actually better than the beginner? I ran both over 16,926 real EURUSD 5-minute bars from June 3 to August 26, 2026.

What exactly was compared?

Every rule below produces a direction on a bar, and every rule is then measured the same way: hold 24 bars, two hours, and count the move in pips in the signal direction before spread. A new signal is skipped while a trade is open. The rules:

• Classic divergence — price makes a lower low while RSI(14) makes a higher low, taken five bars later at the moment the swing point is confirmed. The full method is in my RSI divergence backtest.
• RSI exit from 30/70 — the textbook signal. Buy when RSI crosses back up through 30, sell when it crosses back down through 70.
• RSI simply beyond 30/70 — the impatient version: long while RSI is under 30, short while it is over 70, no waiting for the cross back.
• RSI exit from 20/80 — the same textbook rule with stricter levels, which is the standard advice when 30/70 fires too often.
• Hidden divergence — the continuation pattern: a higher low in price against a lower low in RSI.

Which RSI rule made money?

One rule in five is positive after costs, and it is the one that fires least often. That ordering is not a coincidence, and it is worth taking apart properly.

Why does the textbook 30/70 rule lose?

Not because it is wrong about direction. A 42.7% win rate at a two-hour hold is poor but not catastrophic, and the average loss per trade is only half a pip. The rule loses because it fires 436 times in under three months and each of those round turns costs about a pip on EURUSD, as measured in my spread and swap breakdown. Four hundred and thirty-six near-coin-flips at a pip each is a slow, guaranteed transfer to the broker.

There is a second problem the table shows quietly. Waiting for the cross back through 30 did worse than simply being long while RSI sat under 30 — 42.7% against 46.2%. The patient version has fewer signals and a lower win rate, which means the cross-back is not adding information; it is adding delay. On a 5-minute chart the exhaustion, if there is any, has already happened by the time RSI climbs back over the line.

Does tightening the levels to 20/80 fix it?

It half-fixes it. Moving from 30/70 to 20/80 cuts the signal count by 44%, lifts the win rate by nearly four points and flips the average from -0.53 to +0.35 pips. The direction of travel is unmistakable: the fewer and rarer the signals, the better each one is. Divergence sits at the end of that same line — it is simply the strictest filter of the group, demanding two swing points and a disagreement between them rather than one number crossing one level.

But +0.35 pips a trade is not a strategy. It is roughly a third of the spread. The 20/80 rule is the most seductive row in the whole test because it looks profitable and is not, and it is the same trap I found with tighter thresholds in my stochastic overbought study, where 1,075 extreme readings produced a similar mirage.

What about hidden divergence?

Hidden divergence is sold as the trend-following sibling: price makes a higher low while the oscillator makes a lower low, supposedly marking a pullback inside a continuing move. It produced almost exactly as many signals as the classic version — 95 against 104 — and worse results in every column: 42.1% wins against 50.0%, and negative once you pay the spread.

The interesting part is that the same structural logic applied to the same data gives opposite results depending on which way you read it. That points at something I have now seen repeatedly on this dataset: on 5-minute EURUSD, reversal readings beat continuation readings. It is exactly what happened when I filtered Parabolic SAR by trend strength — the trend-following interpretation lost and the mean-reverting one paid.

Does divergence work better with or against the trend?

The same conclusion arrives from a third direction. Divergence taken against the 200-EMA — the trade every trend-following guide tells you to skip — was worth almost five times more per trade than the trend-aligned version. That is what a reversal signal is supposed to do, and it is the strongest argument in this whole comparison that divergence is measuring something real rather than reflecting the drift of the market. Compare that with the ADX and Supertrend filters, which improved almost nothing they were attached to.

So is divergence worth the upgrade?

On the evidence, yes — with two caveats that matter more than the win rate.

• It is the only positive rule after costs, and the margin over the textbook rule is 708 pips across three months of one pair.
• It is 4.7 times rarer. 92 signals against 436. If the reason you use RSI is that you want something to do, divergence will not give you that.
• Its edge depends on holding. At half an hour divergence itself is negative after spread; the +1.45 requires two hours and the best numbers require four to eight.
• The sample is small. Ninety-two trades supports a conclusion about ranking, not a promise about next month.

Frequently asked questions

Is RSI divergence better than overbought and oversold levels?

On 16,926 EURUSD 5-minute bars, yes and clearly. Divergence made +1.45 pips per trade across 92 signals; the standard exit-from-30/70 rule lost 0.53 pips per trade across 436. After a one-pip spread, divergence is the only version of RSI in this test that ends positive.

Should I use 20/80 instead of 30/70?

It is a genuine improvement — the win rate rises from 42.7% to 46.5% and the average trade flips positive — but +0.35 pips does not cover a one-pip spread, so the rule still loses money in practice. Tighter levels reduce the damage rather than create an edge.

Why does hidden divergence perform worse?

Because it is a continuation signal, and on this dataset the 5-minute EURUSD market rewarded reversal readings and punished continuation ones. Hidden divergence produced 95 signals at 42.1% wins and finished negative after costs, while classic divergence on the same bars won 50.0%.

Does RSI period change the comparison?

Not the ranking. Divergence stayed ahead of the level rules with RSI 7, 14 and 21, though the trade counts shift; the period comparison for plain RSI is in my RSI(7) against RSI(14) scalping test and the divergence grid is in my MT4 setup guide.

Is RSI or stochastic better for these signals?

They behave alike and fail alike at extremes. My stochastic against RSI scalping test ran that comparison on the same bars, and neither oscillator’s raw levels cleared the spread.

What data was this run on?

16,926 EURUSD 5-minute bars, June 3 to August 26, 2026 — the same set behind my Ichimoku comparison and every other backtest in this series, so the numbers are directly comparable across articles.