RSI Divergence Trading Strategy for ES and NQ Futures

RSI divergence looks obvious on a finished chart and becomes ambiguous the moment you try to trade it live. Price makes a new extreme, RSI doesn't, and the story says momentum is fading. But in a strong trend RSI can diverge several times while price keeps going. So the useful question is how to define it so you can test it, and what has to happen before you enter.

What RSI measures

RSI compares average gains with average losses over a lookback, usually 14 periods with Wilder smoothing: RSI = 100 minus 100 divided by (1 + RS), where RS is average gain divided by average loss. It's bounded between 0 and 100. It measures recent price changes, not the amount of buying or selling in the order book. That's worth remembering, because a 'weakening buying pressure' narrative goes beyond what the formula tells you.

The two types

Bearish divergence: price makes a higher high while RSI makes a lower high. Bullish divergence: price makes a lower low while RSI makes a higher low. Either one only says the latest push had smaller momentum than the previous push. It says nothing about when, or whether, price reverses.

Rules that make it testable

Write these down before you look at any chart:

1. Both extremes must be real swing points, defined objectively, such as a high with a set number of lower bars on each side on the 5-minute or 15-minute chart. Two consecutive up candles don't count.
2. RSI at the first swing should be extended by your own definition. Many traders use above 65 or 70 for bearish and below 30 or 35 for bullish. Those are conventions, not proven cutoffs.
3. Optionally require a structural reference near the second extreme, such as a prior session high, weekly extreme or a fixed reference level like KPL. Treat that as a filter to test, not as proof of a reversal.
4. Wait for a trigger. For a bearish setup, that could be the first break of the prior minor swing low on a 2-minute chart after the second high forms.

A worked example (hypothetical)

ES makes a swing high at 5,200 with RSI at 74. After a pullback it makes a second high at 5,206 with RSI at 66. A 2-minute lower low triggers a short at 5,201. The stop goes above the second high with a two-tick buffer, at 5,206.50. Risk is 5.50 points, or $275 per ES contract and $27.50 per MES. A target at 5,190 offers 11 points, which is a 2:1 reward-to-risk ratio. Subtract commissions and slippage, and remember a stop can fill worse than 5,206.50.

The mirror image for bullish divergence uses a lower low with a higher RSI low, a 2-minute higher-high trigger, a stop below the second low and a target at a prior resistance or VWAP.

Where it breaks down

The common failures are predictable. In a persistent trend, repeated divergences get run over one after another. RSI can stay in overbought territory for a long stretch, so entering on the reading alone is just guessing. Divergence drawn across micro swings on a 1-minute chart produces constant signals with no way to tell them apart. Scheduled news can overwhelm any momentum read. And without a trigger, you're entering a warning as if it were a signal.

Timeframes and settings

Higher timeframes give fewer signals that cover larger point ranges, which means wider stops and fewer contracts at the same dollar risk. That's a trade-off you can see in the numbers, not a proven reliability ranking. For settings, 14 is the default. Shorter periods flag more divergences, longer ones fewer, and I'd pick one and keep it across your tests so the journal compares like with like.

How to test it

Log every qualifying divergence, not only the ones that worked: date, timeframe, RSI values, structure nearby, trigger, entry, stop, target and result in R. Then run the numbers with costs. Thirty trades isn't a statistically proven edge, and fifty simulated trades isn't live proof, since simulated fills don't capture queue position or slippage. Compare the setup with and without each filter to see whether the filter earns its place. The edge calculator is a quick way to check expectancy from your win rate and average win and loss. If you use fixed reference levels as a filter, the KPL page explains how they're anchored to the regular-session open, so you know what you're testing against.