Free research · RSI divergence

Does RSI Divergence Work on Gold? We Tested It on 9 Years of XAUUSD

Short answer: no. Across 96 versions on every chart from 1-minute to weekly, none passed our bar. Win rates ranged from 38.4% (25,722 trades) to 58.4% (329 trades).

Tested onGold (XAUUSD)
DataJanuary 2017 to September 2026 (9 years 8 months)
Rule versions tested96
Win rate we measured (after costs)43.7% over 23660 trades (M1 bearish → sell, the biggest sample)Across every chart in the main table: lowest 43.7% on M1 bearish → sell (23660 trades), highest 60.0% on D1 bullish → buy (10 trades). Winning more often is not the same as making money; the tables below show both.

Short answer: no. Across 96 versions on every chart from 1-minute to weekly, none passed our bar. Win rates ranged from 38.4% (25,722 trades) to 58.4% (329 trades).

Read this first, what this test can and cannot tell you

What we tested

Regular RSI divergence, modelled on the settings of TradingView's built-in RSI Divergence indicator: RSI 14; turning points on the RSI with 5 candles each side; the previous turning point 5 to 60 candles back. Bullish: price makes a lower low while RSI makes a higher low → buy at the next candle's open. Bearish: price makes a higher high while RSI makes a lower high → sell. Also tested with an "extreme" filter (RSI below 30 / above 70 at the first point). Exits after 4, 12 or 24 candles. Charts: M1, M5, M15, M30, H1, H4, D1, W1.

Results by chart (default settings, exit after 12 candles)

Chart Direction Trades Win rate Average per trade (bp, after spread) Verdict
M1 bullish → buy 21848 44.2% -0.87 no edge (not significant)
M1 bearish → sell 23660 43.7% -0.89 no edge (not significant)
M5 bullish → buy 4243 48.1% -0.45 no edge (not significant)
M5 bearish → sell 4800 46.7% -1.05 no edge (not significant)
M15 bullish → buy 1364 51.0% -0.07 no edge (not significant)
M15 bearish → sell 1591 46.7% -3.47 no edge (not significant)
M30 bullish → buy 664 55.1% +2.22 no edge (not significant)
M30 bearish → sell 804 50.1% -1.05 no edge (not significant)
H1 bullish → buy 329 58.4% +3.64 no edge (not significant)
H1 bearish → sell 454 45.2% -7.43 no edge (not significant)
H4 bullish → buy 86 52.3% +27.26 too few trades to judge
H4 bearish → sell 122 48.4% +9.84 too few trades to judge
D1 bullish → buy 10 60.0% -41.66 too few trades to judge
D1 bearish → sell 23 56.5% -34.45 too few trades to judge
W1 bullish → buy 0 , , too few trades to judge
W1 bearish → sell 7 57.1% -45.19 too few trades to judge

With a stop and a target (stop beyond the divergence, target twice the risk)

A coin-flip market reaches a 2:1 target before the stop about one time in three. That is the number to beat.

Chart Direction Trades Won (target before stop) Chance would give Average result (R)
M1 buy 21687 24.9% 33.3% -0.24 R
M1 sell 23136 22.4% 33.3% -0.32 R
M5 buy 4121 29.0% 33.3% -0.12 R
M5 sell 4660 27.1% 33.3% -0.18 R
M15 buy 1323 31.2% 33.3% -0.06 R
M15 sell 1568 28.5% 33.3% -0.15 R
M30 buy 658 34.9% 33.3% +0.06 R
M30 sell 786 29.6% 33.3% -0.11 R
H1 buy 319 36.9% 33.3% +0.11 R
H1 sell 427 28.9% 33.3% -0.13 R
H4 buy 85 38.8% 33.3% +0.16 R
H4 sell 118 33.6% 33.3% +0.01 R

Year by year: bullish divergence, 1-hour chart, exit after 12 candles

Year Trades Win rate Average per trade (bp, after spread)
2017 36 47% -3.59
2018 35 51% +0.58
2019 33 70% +16.83
2020 28 50% -11.11
2021 36 64% +5.83
2022 46 59% +7.07
2023 43 51% -2.44
2024 26 85% +25.61
2025 22 64% +25.30
2026 24 50% -24.59

Does the divergence itself add anything?

We compared every divergence with the same setup where price made the new extreme but RSI did not diverge. From the 15-minute to the 1-hour chart the difference was not significant. On the 4-hour chart bullish divergence did better (+23.1 bp, t +2.63), but on only 92 trades and as one of 24 comparisons, a lead worth watching, not a finding.

How we tested

Think we got it wrong? Show us. Send your rules, your data source and your result (form fields: your definition · timeframe · data source and dates · costs you charged · your result and sample size · a link or file). If your test holds up, we publish the correction on this page, dated, with your name if you want it.

Request a backtest

Want a concept, a variation or another market tested the same way? Tell us what to test and where to send the result. It is free.

We only use your contact to send you this result.

Disagree? Bring results.

We publish numbers, not opinions, and we answer with numbers. If you believe a concept works and our test missed it, send us your exact rules, the market and timeframe, your data source and dates, the costs you charged, and your result with its sample size. We will run your version and reply with the actual statistics: every trade, the win rate, the result after costs, year by year. If your test holds up, we correct this page, dated, with credit to you if you want it.

Claims without results will not change a page. A screenshot of a winning trade is one trade; the tables here are thousands.

Send it to support@fxsynapseai.com with the subject “Research correction”, or use the request form above.