Where Does a Good-Looking EA Backtest Go in Real Conditions? 60 Textbook Strategies on 11 Years of Gold
- We tested the moving-average crossover (EMA crossover), a textbook strategy at the core of many EAs, in 60 variations on real hourly gold prices, 2015–2026.
- Costs eat the profit: the best variation earned only $1.40 per trade before costs. At $1.50 of spread plus slippage per trade, 11 years of profit becomes a loss.
- The period you show changes everything: the best 9 months (Jun 2025 – Feb 2026) made 83% of the profit of the whole 11.75 years, while 2015–2019 combined lost money.
- Every short-target scalping variation lost money, even with zero costs.
Why run the experiment ourselves
In How to Read an EA Backtest Report we explained what to look for in a report. This article works from the other side: we build the backtests ourselves and show how many ways the same system can be made to look good or bad, without faking a single number.
We chose a strategy everyone knows and that is not a secret, a fast moving average crossing a slow one, so anyone can reproduce it. (This article does not test the logic of any particular EA, including ours.)
Method
| Data | Hourly XAUUSD, Jan 2015 – Sep 2026 (11.75 years) |
| Signal | Fast EMA (5, 8, 10, 13, 20, 30) crossing slow EMA (50, 80, 100, 150, 200): 30 pairs · entry at the next candle's open · one position at a time |
| Swing version | SL = 2 × ATR · TP = 3 × ATR · or exit on the opposite cross |
| Scalp version | SL = 0.5 × ATR · TP = 0.5 × ATR · or exit on the opposite cross |
| Candle touches both SL and TP | Counted as SL first (pessimistic) |
| Standard cost | $0.30 per trade (a typical Strategy Tester spread setting) |
| Units | Profit in dollars per 0.01 lot (multiply by 10 for 0.10 lot) |
| Optimise / test | Best settings chosen on 2015–2020, then checked on 2021–2026, which was not used for optimising |
Finding 1: most textbook variations don't make money
In the optimisation period (2015–2020), only 4 of 30 swing variations were profitable. None of the scalp variations made money, in either period.
The best scalp variation lost $0.29 per trade on average before costs. After real costs it loses more. That is worth remembering when an EA advertises "small, frequent profits" on gold: the smaller the target per trade, the larger the share of it that costs take.
So if an EA claims to use just a crossover or a single basic indicator and shows beautiful profits every year, ask how its settings and test period were chosen.
Finding 2: cost per trade decides everything
The best swing pair was EMA 20 crossing EMA 80, trading about 81 times a year with an average gross profit of $1.40 per trade. Here is how its 11.75-year total changes with the cost per trade:
| Cost per trade (spread + slippage) | 11.75-year total | vs zero cost |
|---|---|---|
| $0.00 (no costs) | $1,334 | 100% |
| $0.30 (typical tester setting) | $1,047 | 78% |
| $0.60 | $760 | 57% |
| $1.00 | $378 | 28% |
| $1.50 | −$100 | loss |
Same system, same settings, same data. Only the cost per trade differs, and the result ranges from a nice profit to a loss. Real costs depend on the broker, account type, commission and timing; around news, gold spreads can widen several times (see Which News Really Moves Gold?).
How to use this: divide the report's net profit by the number of trades. If the result is close to your broker's real spread, the system is hanging by a thread.
Finding 3: pick the period, change the picture
This strategy made $1,047 over 11.75 years. Look at where it came from:
- 2015–2019: five years combined, a $87 loss
- 2020–2024: five years combined, $118 profit (including a $269 year and a $172 losing year)
- 2025 – Sep 2026: just over one year, $1,017
The best 9 months (June 2025 – February 2026) made $869, or 83% of the total. Show only that window and the strategy looks brilliant. Show only 2015–2019 and it looks worthless. It is exactly the same system.
The reason: trend strategies make money when the market runs one way for a long time, and gold rallied hard through 2025–2026. Any strategy looks smart in a market that suits it. The question to ask is how much it loses, and whether it survives, when the market doesn't suit it.
Finding 4: optimising guarantees nothing
We picked the best settings on 2015–2020 and looked at 2021–2026, which was not used for optimising:
- In this experiment the best past pair (EMA 20/80) happened to stay the best in the new period too. We'll say plainly that this was better than we expected.
- But of the 5 best pairs in the past, only 2 were still profitable in the new period. The other 3, which looked good during optimisation, lost money.
- And the best pair's profit in the new period ($865) came entirely from the last year or so of strong gains; 2021–2024 combined lost $152. It did not come from consistency.
The lesson: an optimised backtest is the upper bound of what might happen, not what you should expect. A credible EA should show results for the optimisation period and the unseen period separately, and be profitable in both.
Summary: 4 questions for every EA backtest
- What is the profit per trade, and does the tested cost match reality? If profit per trade is close to the real spread, lower your expectations sharply.
- How was the test period chosen? Ask for every year, not just the best stretch, especially the years when the market did not suit the strategy.
- Are there results for data not used in optimisation? If only optimised results exist, those numbers are an upper bound.
- For small-target scalping systems, is there live-account evidence? These systems are the most cost-sensitive; backtests mean little without live results.
If you have a report file, try our free EA checker.
Limitations
- This is the simplest textbook strategy, used to show how costs, period and optimisation affect results. It does not mean more sophisticated strategies behave the same way.
- We used hourly candles, so moves inside the hour are approximated pessimistically (SL before TP).
- Cost per trade is a fixed number; in reality spreads vary with time of day and market conditions.
This article is for educational purposes and is not investment advice. Trading gold and CFDs carries a high level of risk and you may lose all of your capital. Past performance does not guarantee future results.
