- Take Wilder's RSI of the close over a chosen period.
- When RSI closes below the oversold line (or, in the other version, back above it), buy at that bar's close — either the bar that closes below it, or the bar that closes back above, whichever version you were taught.
- Close the trade when RSI reaches the overbought line — and, if you are trading both sides, the mirror image for shorts.
- Flat at the end of the session. One position at a time.
- Our choices, not anyone's rule — nobody owns this strategy, so these had to be picked: flat at the end of the session; 5-minute bars; New York hours (9:30–16:00) for gold too; a setting needs at least 60 trades to be scored. The RSI runs on one continuous chart, as on a trading screen — it does not restart each morning.
- Costs are real: $3.00 a round turn on Nasdaq micros, $25.00 on gold.
- RSI at any bar uses closes up to and including that bar. Nothing looks ahead.
| trades | win | a trade | random, held as long | the rule is worth | |
|---|---|---|---|---|---|
| Nasdaq · the famous RSI 14 30/70 | 493 | 51.5% | −$0.87 | $6.27 | — |
| Nasdaq · the highest win rate we found | 3,604 | 61.2% | −$0.59 | — | — |
| Nasdaq · best of 47 (RSI 14 25/75 reclaim long) | 359 | — | $26.00 | $5.08 | $20.93 |
| Gold · best of 48 (RSI 14 25/75 touch long) | 460 | — | $40.67 | −$45.52 | $86.20 |
The famous numbers lose money. RSI 14 with 30/70, long only, is −$0.87 a trade over 493 trades. Those are the defaults on every platform and the numbers nearly every course teaches.
A high win rate is not an edge. The best win rate we found is 61.2% — right almost two times in three — and it still makes −$0.59 a trade over 3,604 trades. Waiting for an oversold reading gets you into a lot of small winners and a few large losers, because nothing stops the fall while you wait for RSI to climb back.
Every setting, not the one that suits us. The table below changes with the market and the side rule you have selected. 21 of 47 made money on Nasdaq; 3 of 48 did on gold — and not one of them clears the bar once you count how many settings were searched.
No pattern that holds on both markets. Looser lines are right a little more often, but the money does not follow them in a straight line, and waiting for the reclaim instead of buying the touch helps on Nasdaq and hurts on gold.
Everything else worth saying
The control is the whole test. Any rule that holds a long for a while on a market that rose for four years will make money without doing anything clever. So each setting is measured against entering at a random bar and holding for the same number of bars, with the same costs and the same side rule. If the oversold reading cannot beat that, it is measuring the market, not the signal.
The bar is set for the whole search, not each cell. Trying 95 settings and keeping the best is how noise gets published, so the threshold is divided by the number of settings tried. Nothing here comes close to it either way.
Costs weigh most on the settings that fire most. At $3.00 a round turn on Nasdaq and $25.00 on gold, a rule has to be right by more than the commission before it is worth anything.
What this test does not do
We tested RSI as it is taught — oversold, buy; overbought, out. That is the rule beginners are given.
- No filter, no context. Most people who use this add something: only trade with the higher-timeframe trend, only in certain hours, only with a stop. Any of those could change the result, and none of them are part of the rule as taught.
- No stop or target. The trade runs until RSI reaches the opposite line or the session ends, which is how the rule is usually described. That is also its weakness: nothing stops the fall while you wait.
- Intraday, five-minute bars. Daily RSI is the other common version. Four years of sessions gives too few daily signals to score, so it is not covered here.
- No divergence. Plenty of traders use RSI for divergence rather than for the 30 and 70 lines. That is a different rule and is not tested here.
- Two markets. More would make the result stronger.
- The replay is a sample. Every number is scored over all 1,047 Nasdaq and 1,663 gold sessions; the chart ships the most recent 1047 of each so the page stays light.
If you use a pair that is not in the grid, or a filter you think matters, say so and we will test it the same way and publish it.
Nobody owns this one. It is on every platform by default and in every beginner course, which is exactly why it was worth checking.