- A backtest is a hypothesis. Trust a live record you can check, and treat a perfect equity curve as a question, not an answer.
- On gold, the danger is how losses are handled. Averaging down and growing position size can look smooth for months, then lose most of an account in one move.
- Test the EA at your own broker and your own size before you rely on it. Spread, slippage and contract size change the result.
To evaluate a gold EA, check seven things: whether there is a live record you can check; whether it has enough time and trades; how deep its drawdown goes on equity; how it handles losing positions; how its backtest was made; whether it still works at your broker, spread and size; and what limits stop it when it is wrong. An EA that fails the second or the fourth check is a warning even if its profit line looks perfect.
An EA (Expert Advisor) is a program that trades automatically on MetaTrader. If the term is new, start with what is an Expert Advisor in MT5? Gold deserves extra care. It moves fast, spreads widen around data releases, and a long one-way trend can trap a strategy that averages down. A smooth curve on gold often hides a risk that takes months to show.
1. A live record you can check
Ask for a real account, with the broker and server visible, and a record on a site the seller does not control. How to read a Myfxbook record shows what to look at. Demo results, screenshots and backtests are not a live record. A seller who cannot show one has already answered this check.
2. Enough time and enough trades
A few weeks of profit says little. Look for at least 26 weeks, and enough trades that no single trade explains the result. A record that skips the hard weeks, or that starts right after a new account was opened, deserves suspicion.
3. Drawdown, on equity
Read the maximum drawdown beside the return, and find out whether it is measured on balance or on equity. On balance, an EA that holds losing positions can look smooth while its floating loss grows. See what is drawdown and why Myfxbook and MT5 can disagree.
4. How losing positions are handled
This is where most gold EAs that fail, fail. Look in the trade list for three patterns: size that grows after a loss, several positions opened against a move, and losing trades held far longer than winning ones. They produce small, frequent wins, and one large loss.
The next table is an illustration, not a real EA. It assumes a 100-ounce contract, a $10 move against each trade, and a lot size that doubles after every loss.
| Trade | Lots | Loss on a $10 move | Total loss so far |
|---|---|---|---|
| 1 | 0.10 | $100 | $100 |
| 2 | 0.20 | $200 | $300 |
| 3 | 0.40 | $400 | $700 |
| 4 | 0.80 | $800 | $1,500 |
| 5 | 1.60 | $1,600 | $3,100 |
| 6 | 3.20 | $3,200 | $6,300 |
On a $5,000 account, each of the first five trades looks small. Five losses in a row, which a strong gold trend can produce, cost $3,100, and the sixth trade would need more than the account has left. Ask every seller the same question: what happens after the fifth loss?
5. Gold EA backtesting that deserves trust
A backtest is a hypothesis about the past. Before you give it any weight, ask:
- Which tick data? In MetaTrader 5, “Every tick based on real ticks” uses recorded ticks, while “Every tick” generates them. Generated ticks flatter strategies that depend on fast moves.
- Which costs? Spread, commission and swap must be included, and spread should widen at news times.
- Which period? Was the test run on data the settings were not tuned on? A parameter set optimised on the same data it is tested on is a description of the past, not a test.
- How many settings were tried? The more settings, the more likely that one looks good by luck.
6. Does it work at your broker and your size?
An EA tuned on one broker’s prices can fail on another’s. Check the symbol specification at your broker: the contract size (usually 100 ounces for XAUUSD, but not on every account type), the minimum lot, the typical spread, the swap, and the trading hours. If the EA scales its size to your balance, check that the smallest trade it would send is not below your broker’s minimum. Gold (XAUUSD) copy trading covers sizing in detail.
7. Limits that stop it when it is wrong
Ask what each position, or group of positions, does when it goes wrong. Is there a defined exit? Is total exposure capped? Is there a daily loss limit, a spread filter, a pause around major news? What happens if the connection drops? An EA that cannot answer these is relying on the market to cooperate.
Putting it together
Mark each check as pass, unclear or fail. A fail on checks 1, 2 or 4 is a reason to walk away. For anything unclear, run the EA on a small live account for several weeks and compare its results with what the seller showed. The same logic applies to a strategy you copy rather than run yourself: see how to choose a copy trading master and does AI trading work?.
Questions
Are backtests useless?
No. A backtest shows whether an idea behaved sensibly on past data and what it risked. It cannot show that the idea will work live, so it is a starting point and not proof.
What is a martingale EA?
An EA that increases its position size after a loss to recover faster. It tends to win often and small, and it can lose most of an account in one losing run.
How long should I test an EA before trusting it?
Several weeks on a small live account at least, and ideally a live record of 26 weeks or more before you give it real size. A demo account hides fills and costs, so a small live account tells you more.
This is general information, not investment advice. Leveraged products can lose more than you deposit. Read the full risk disclosure