A MetaTrader 5 indicator for gold: what a 6-year backtest shows
Most indicators sold online show an equity curve that goes up in a straight line. Here, we do the opposite: these are all the numbers of our method on gold, including what does not look good.
The method tested
Hybrid Or combines three timeframes (details in our article on multi-timeframe analysis):
- D1: trend, using the EMA 20 and EMA 100;
- H4: pullback, with the RSI dropping below 48 (buy) or rising above 52 (sell);
- H1: entry on a break of the high or low of the last 20 candles.
Two quality filters are added (version 2.1 of the indicator):
- no signals at night, between 10 pm and 3 am (MT5 server time), when liquidity is thin and false breakouts are common;
- a pullback of at least 38.2% of the last H4 swing (a Fibonacci level): we only enter after a real pullback, not a mere pause.
Every trade has a stop at 1.5 × the H4 ATR, a partial take-profit of 67% at 1 R, then the stop moved to the entry price, and a final target at 2 R.
How we tested it
- Tool: the MetaTrader 5 strategy tester.
- Data: 1-minute XAUUSD data from the broker XM, from January 1, 2020 to September 24, 2026.
- Costs: spread and overnight fees (swap) included.
- Risk: 1% of capital per trade, one trade at a time, starting account of $10,000.
Above all, we kept the periods separate to avoid "cheating" with the past:
- the settings were chosen on 2020–2024;
- they were then checked on 2025, without being changed;
- and one last time on 2026, a period that was never used to choose anything.
The results
| Metric | Value |
|---|---|
| Positions | 156 (including 40 sells), about 23 per year |
| Winning positions | 56.4% |
| Profit factor (gains ÷ losses) | 1.51 |
| Maximum drawdown | 7.2% |
| Winning months | 45 out of 81 (24 losing, 12 with no signal) |
| Total result | +$3,619, or +36.2% over 6 years and 9 months |
Result by year:
| Year | Result |
|---|---|
| 2020 | −$322 |
| 2021 | +$523 |
| 2022 | +$905 |
| 2023 | +$178 |
| 2024 | +$844 |
| 2025 | +$857 |
| 2026 (to September 24) | +$634 |
What to take away, honestly
- The method has an edge, but a modest one. About 5% per year on average with 1% risk per trade. That is a long way from promises to "double your account".
- There are losing periods. 2020 ended in the red (−3.2%), and about 3 months out of 10 are losing months.
- The maximum drawdown stays contained (7.2%), thanks to the filters, the fixed 1% risk and a systematic stop.
- Carefully selected signals. About 23 per year, or one every two weeks on average: patience is required.
- The validation periods hold up. 2025 and 2026, never used to choose the settings, remain positive.
What did not work
We tested more than 2,000 combinations before getting here. A few lessons:
- A single timeframe was not enough. No pure H1 variant held up over both periods: at best, they were barely profitable over 2020–2024, with drawdowns of around 40%.
- Optimizing is dangerous. In a first test on EUR/USD, the best settings for 2023–2024 lost money in 2025–2026: the correlation between the results of the two periods was close to zero. Hence the strict separation of periods for gold.
- Chasing a high win rate at all costs is a trap. Some filters boosted the win rate on past data, then collapsed in 2025–2026.
- Fibonacci alone was not enough. Used on its own, the Fibonacci filter shone over 2020–2024 but lost money in 2025–2026. It was only kept in combination with the night filter, because that combination held up over both periods.
- First version, without filters: +29.5% over the same period, but with two losing years (2021 and 2023), 53% winning positions and an 11.5% maximum drawdown. The current version earns more (+36.2%), with much smaller drawdowns.
- Chasing a signal every week does not work. A faster engine (H4/H1/M15) that gave 1 to 2 signals per week lost money once costs were included. We dropped it.
The limits of a backtest
A backtest is not a promise. It does not perfectly reproduce slippage, weekend gaps, connection outages or human error. Data differs slightly from one broker to another, so signals with your broker may be slightly different from those in the backtest (XM data).
Past performance, whether actual or simulated, is not indicative of future results. Only trade money you can afford to lose.
If you want to follow the method live, the Hybrid Or indicator displays the same signals on MetaTrader 5, with the stop-loss and take-profit levels. And the result of every signal is published, winners and losers alike.