Forex Trading Systems Tested: 1 of 20 Runs Passed, None on the 5-Minute Chart, and RSI-2 Won 58% of Its Trades While Losing

Most forex trading systems are sold as a set of rules: a trend filter, an entry, a stop, a target. I took five that are taught everywhere, coded every rule, and ran them on the same data. On the 5-minute chart none of the five made money in a way that held up. One combination did on the 15-minute gold chart: buying a close below the lower Bollinger Band with RSI under 30, and selling the mirror image. Here are the numbers for all 20 runs.

Which five trading systems did I test?

ATR is the 14-bar Average True Range of the chart traded. Systems 1 to 4 also close any trade still open after 48 bars.

How was each system tested?

• Data: EURUSD M5 from June 3 to August 26, 2026 (16,926 bars) and gold M5 from June 16 to August 26, 2026 (13,736 bars). M15 bars were built from the same data.
• One trade at a time: a new signal while a trade is open is skipped, the way a person or an EA would trade it. Entry is at the close of the signal bar. If a bar touches both stop and target, the stop counts first.
• Cost: 1 pip per EURUSD trade and $0.30 per ounce of gold, as in my spread and swap cost test.
• Three checks: positive after cost; positive in both halves of the data (split at July 15); and above the best 5% of 300 runs of random entries with the same exits and the same number of trades. For the systems with a trend filter, random entries were also taken only on the filter’s side of the market, so the filter alone could not explain the result.

One note on the EURUSD data: prices in it move in steps of about 1.3 pips. That is too coarse to judge sub-pip edges, so I only read EURUSD results larger than that.

What did each system make on the 5-minute chart?

Win rate counts trades positive after cost.

On EURUSD M5 the two trend systems lost before the spread was taken. The two mean-reversion systems, RSI-2 and Bollinger with RSI, made a quarter of a pip a trade before cost, a quarter of what the spread takes. The London breakout made money over the whole test, but all of it in the first half.

On gold M5 four systems lost after cost. The London breakout made +$11.06 an ounce a trade in both halves, which looks like the best result in the test. The next sections explain why I do not count it.

Why did RSI-2 lose with a 58% win rate?

RSI-2 is the system with the best win rate in this test and one of the worst totals. It buys a sharp dip and sells as soon as price closes back above a 5-bar average. That exit takes small wins quickly. The losers run until the 2 ATR stop or until price finally recovers, so an average loser was about two and a half times an average winner on EURUSD. A 58% win rate needs winners at least 72% the size of the losers to break even. RSI-2’s winners were about 41%.

The trend systems had the opposite shape: about one winner in three, with winners bigger than losers, and still a loss. A high win rate or a 1:2 target on its own tells you nothing about whether a system pays.

Did any system beat random entries?

Random entries use the same stop and exit rules and the same number of trades. For the London breakout they are random bars in the same 07:00 to 10:00 UTC window on the same days, in a random direction.

The MA trend system on EURUSD M5 did slightly worse than random entries with the same stop and target. Its entry added nothing.

The London breakout on gold is the warning case. Gold rose $323 an ounce over the test, from $4,352 to $4,675, and the breakout’s 11 long trades made +$24.07 each while its 14 shorts made +$0.83. Random trades placed in the same morning window, with the same stop at the range and the same 16:00 close, made +$3.37 in a typical run and up to +$12.66 in the best 5%. With 25 trades, the system sits inside that range. It may have an edge, but this data cannot separate it from a rising gold price and luck.

Bollinger with RSI was the only system that beat random entries on every one of the four charts. On EURUSD M5 its edge was a quarter of a pip, too small to survive the spread. On gold M15 it was large enough.

Which system passed all three checks?

The one run that passed was Bollinger Bands 20/2 with RSI 14 at 30/70 on gold M15: 114 trades, +$1.69 an ounce a trade after cost, +$1.53 in the first half and +$1.78 in the second, 37.7% winners, profit factor 1.29, worst drawdown $156.70 an ounce. Its longs made +$0.77 a trade and its shorts +$2.60, so this was not just gold going up. The best 5% of random entries reached +$0.33.

To see whether that was one lucky setting, I ran 54 variants: band periods 15, 20 and 25, widths 2 and 2.5, RSI at 30/70, 25/75 or off, and stops of 1, 1.5 and 2 ATR. On gold M15, 49 of 54 were positive after cost and 29 were positive in both halves with at least 30 trades. On EURUSD M5, 0 of 54 were positive. The full setup is in my Bollinger Bands and RSI system guide for MT4.

What should you take from this before buying or building a system?

• Ask for the result after cost, by half, against random entries. A system that cannot show those three is a story. My algorithmic rule vs random entry test shows how often random entries look as good as a rule.
• Do not judge by win rate. RSI-2 won 58% of EURUSD trades and lost money. Look at the size of the average loser next to the average winner.
• Test the market before the system. The same Bollinger entry beat random on all four charts, but only on gold M15 was the edge larger than the cost. My gold vs EURUSD test with one rule shows the same split.
• Distrust a great result on a few dozen trades. The London breakout’s +$11 an ounce on 25 trades fell inside what random morning trades did. The optimized vs default settings test shows how small samples fail later.

Which part of a system does the work, the filter, the entry or the exits, is the subject of my trading system vs its own entry signal comparison. For the single indicators behind these systems, see the MACD crossover backtest and the Bollinger Bands vs RSI test.

Frequently asked questions

What is the most profitable forex trading system?

In this test of five popular systems on EURUSD and gold from June to August 2026, the only one that passed all three checks was Bollinger Bands 20/2 with RSI 14 at 30/70 on gold M15: +$1.69 an ounce a trade after cost on 114 trades. That is 10 weeks of one metal, not a guarantee.

Do forex trading systems work on the 5-minute chart?

None of the five did in this test. On EURUSD M5 four lost after a 1-pip spread and the London breakout, the fifth, lost 3.29 pips a trade in the second half. On gold M5 no system passed the halves and the random-entry check together.

Does the RSI-2 strategy work in forex?

Not on EURUSD or gold M5 here. It won 58% and 61% of trades, but its average loser was much larger than its average winner, and it lost 0.72 pips and $0.24 an ounce a trade after cost.

Does the London breakout strategy work?

On gold it made +$11.06 an ounce a trade on 25 trades, but gold rose $323 in the test and random trades in the same morning window made +$3.37 typically and up to +$12.66. On EURUSD it made +0.32 pips a trade and lost in the second half.

How do I test a forex trading system myself?

Code every rule, take one trade at a time, subtract the spread, split the data in two and compare with random entries using the same exits. My guide to backtesting an EA on MT5 shows how to do it in the Strategy Tester.