What Backtests Don't Tell You: I Tested 6 Popular Gold EAs — Here Is How Not to Be Fooled
🔍 Most gold EAs that look perfect in a backtest stop looking perfect the moment you test them outside the period the seller chose.
Over the last weeks I ran six well-known XAUUSD Expert Advisors from the MQL5 Market through the MT5 Strategy Tester, read their trade lists line by line, and rebuilt their logic on my own 1-minute price data from 2018 to 2026.
⚠️ I am not naming the products. The goal is not to attack anyone. It is to share what the data showed, so that you can check any EA yourself — before you pay for it.
🔬 How I tested
- ✅ Free demo versions in the MT5 Strategy Tester, XAUUSD, default settings unless noted.
- ✅ Fixed 0.01 lot, so compounding does not hide what each trade does.
- ✅ Long periods (2018–2026), and especially the years the seller did not show.
- ✅ One year re-run with real ticks, to check that the tick model was not flattering the result.
- ✅ Each strategy inside a multi-strategy EA tested on its own, so trades from different strategies do not get mixed.
- ✅ Every logic I extracted was re-simulated on 1-minute data with a realistic cost of $0.20–$0.40 per side.
🧩 Lesson 1 — Different names, the same breakout
Two popular EAs from different sellers turned out to trade almost the same trades. In the tester, 99% of one EA's entries happened within 5 seconds of the other EA's entries, and 95% of its pending-order prices were identical. When I shifted the time by three days and repeated the check, the match fell to 0% — so it was not a coincidence.
The common core is simple: pending stop orders at the high and low of the last N days, with a stop-loss and a take-profit. Many long-running gold EAs on the Market are variations of this one idea.
👉 Buying three "different" breakout EAs may give you the same trade three times — and three times the risk.
📅 Lesson 2 — Test the years the seller did not show
One trend-pullback EA showed a smooth equity curve from January 2020 to September 2026 on its product page. I tested only the two years before that, 2018–2019, with the same fixed 0.01 lot:
- 📉 Net result: −$509 on a $1,000 account
- 📉 Profit factor 0.38 — 5 winners out of 51 trades (9.8%)
- 📉 Maximum balance drawdown 52%, with 16 losses in a row
The reason was visible in the settings: a $15 stop-loss with a take-profit six times larger, $90. In 2018–2019 gold rarely moved $90 in one direction within a trade, so the target was almost never reached. From 2020 the moves became bigger, and the same settings started to work.
👉 The settings fit the volatility of the period that was shown — not gold in general.
🗓️ Lesson 3 — Look for months with no trades
The most surprising case was a multi-strategy breakout EA. Its 2018–2026 backtest made a profit in every single year, with a profit factor of 1.48. A one-year re-run with real ticks gave almost the same result, so the tick model was not the issue.
The trade list told a different story. On some days the EA placed no orders at all — and those days were not scattered. They were whole calendar months: 18 months for one strategy and 14 months for another, all between 2018 and 2024. In 2025 and 2026 it did not skip a single month. No market condition I checked (range size, volatility, weekday, the previous month's move) explained the skipped months.
When I ran the same entry and exit rules on every day, the skipped months were exactly the losing ones:
| Strategy | PF in the skipped months | PF in the traded months | PF on all days |
|---|---|---|---|
| Strategy 1 | ❌ 0.54 | 1.30 | 1.14 |
| Strategy 2 | ❌ 0.62 | 1.15 | 1.05 |
My simulation on 1-minute data, $0.20 cost per side.
Nobody can know in advance which month will lose. A backtest that avoids past losing months by date shows the past more kindly than a live account will see the future.
👉 The honest profit factor of this logic was about 1.05–1.14 — not 1.48.
💸 Lesson 4 — Small profits need perfect execution
Several EAs win often but win small. One strategy above closes most winners at about half of its risk — around $2.50 per trade at today's prices. Its result depends almost entirely on what each trade costs:
| Extra cost per side | $0.00 | $0.10 | $0.20 | $0.30 |
|---|---|---|---|---|
| Profit factor | 1.48 | 1.29 | 1.15 | 1.01 |
Breakout orders at the London open often fill a little worse than the order price in a live account. A few cents of slippage can turn this kind of edge into nothing.
A second pattern was edges that live only in fast markets. A pullback scalper I tested earned 75% of its 2020–2023 profit in 2020 alone (profit factor 4.28). In calmer years it traded about 30 times a year, and in 2023 it lost (0.84). Its thresholds were set in fixed dollars, so it traded far more when gold moved more.
👉 A strong recent backtest from such an EA mostly tells you that gold has been volatile.
⚖️ Lesson 5 — Big live growth usually means stacked risk
One EA with an impressive live signal had a 53% drawdown on that signal. I expected martingale, but there was none: every trade in the tester used the same lot. The risk came from somewhere else — several strategies entered the same direction at the same time, up to 12 positions at once, each with a high risk setting.
With a fixed 0.01 lot, the same logic had a drawdown of only about 5–7%. Same entries, same exits — roughly ten times the drawdown, only because of how much was risked at once.
👉 Before you copy a fast-growing signal, check the maximum number of open positions and the risk per position — not just the growth.
🛡️ How not to be fooled — 9 habits
- 🧪 Trust the test you run yourself, not the screenshot. Every Market EA has a free demo. Twenty minutes in the tester tells you more than any image on a product page.
- ⏳ A live signal younger than one year proves little. Check that it runs the same EA, the same version and the default settings you will use.
- 🎯 A 90%+ win rate with a wide stop-loss means many small wins and a few large losses that may simply not have arrived yet.
- 🏷️ "Only 3 copies left" and "price rises every 10 sales" are sales techniques, not signs of quality.
- 🤖 "AI" or "GPT" EAs that need an internet connection cannot be reproduced in the tester. Treat their backtest images as unverifiable.
- 🔢 Check the arithmetic of the growth. "$100 to $400,000" is usually compounding at a high risk, not a better strategy.
- 💬 Read the lowest-rated reviews and the Comments tab, and see whether the author answers problems or only praise.
- 🚫 Check the seller's status. If the page says "Author debarred from selling products", do not expect updates or support.
- 🪙 Start small. Rent for one month if you can, run a demo or the smallest live account first, and compare live results with the backtest of the same weeks.
✔️ Checklist — 8 tests before you buy a gold EA
All of these can be done with the free demo and the MT5 Strategy Tester:
- ☑️ Test with a fixed 0.01 lot, so compounding does not hide the size of each trade.
- ☑️ Test the years the seller did not show. If the product page starts in 2020, run 2018–2019.
- ☑️ Open the monthly report and look for months with zero trades. Whole months without orders need an explanation.
- ☑️ Re-run one year with real ticks and compare the profit factor with the faster model.
- ☑️ If the EA has several strategies, switch them on one at a time.
- ☑️ Look at the average winner in dollars. If it is only a few dollars, ask what $0.20–$0.30 of slippage per trade would do.
- ☑️ Check the yearly results. A profit made mostly in one volatile year is not a steady edge.
- ☑️ On the live signal, check the maximum number of open positions and the risk per trade, not only the growth.
👤 About the author
I develop Gold Compass Pro MAX, an Expert Advisor for XAUUSD. I hold my own EA to the same standard: please run this checklist on it too — the free demo is on the product page, and the user manual explains every setting.
🛒 Product page: Gold Compass Pro MAX on the MQL5 Market
▶️ YouTube: Gold Compass Pro channel
📘 User manual in 14 languages: EN · RU · ZH · ES · PT · JA · DE · KO · FR · IT · TR · ID · TH · MN
All figures come from MT5 Strategy Tester runs and my own simulations on historical data. Past results, backtested or live, do not guarantee future results. Trading gold with leverage can lose money.


