How to judge an Expert Advisor before you run it on real money

An Expert Advisor (EA) is a program that trades for you on MetaTrader. Thousands are for sale, and almost all of them come with a beautiful backtest. We build and test EAs ourselves, and we've rebuilt and stress-tested several popular commercial ones to see how they actually work. The pattern is consistent: the backtest you're shown is the best case, and the questions below are what separate a robust system from a curve-fitted one.

1. How was the backtest made?

MetaTrader 5's strategy tester has several modelling modes. Only "Every tick based on real ticks" uses the broker's actual price history. Faster modes such as "1 minute OHLC" or "Open prices only" guess what happened inside each bar, and for EAs with tight stops, trailing stops or many trades a day, that guess can flatter the results badly.

We've seen a trailing-stop system that looked clearly profitable on one-minute bars and lost money once it was tested on real ticks. The bar data had assumed price moved in the order that suited the trailing stop.

Also check the spread used in the test. A fixed, low spread on a scalping EA can be the difference between profit and loss.

2. Is there an out-of-sample test?

If the EA's settings were optimised on 2020 to 2024 and the backtest shows 2020 to 2024, you're looking at a curve fit. It's guaranteed to look good, because it was tuned to that exact data.

Ask for results on a period the settings were never tuned on: earlier years, later years, or other symbols. A robust EA performs reasonably everywhere, not brilliantly in one window. Several EAs we tested lost all of their edge in the years after the period they were optimised on.

3. Look at equity drawdown, not just balance

The balance curve only moves when trades close. Equity includes open trades. Grid and martingale systems, which open more trades as price moves against them and close everything together at a small profit, show a smooth, rising balance line while equity underneath drops much further.

When we rebuilt a grid-style gold EA and tested it on real ticks, its closed-trade drawdown looked moderate, but its equity drawdown was close to half the account. On a currency-pair grid EA, one strong trend took the account down by more than 90%. The smooth balance curve hid all of it until it didn't.

Always ask for the maximum equity drawdown, and look at the equity curve, not just the balance curve.

4. Warning signs in the trading logic

• No stop loss, or a stop so wide it never gets hit in the backtest.
• Lot sizes that increase after a loss (martingale).
• Many positions opened in the same direction as price moves against them (grid or averaging).
• A win rate above 90% with small wins and occasional very large losses.
• Long gaps between trades that nobody can explain. That often means a filter that happened to avoid the bad days in the backtest.

None of these automatically means the EA will fail. They mean a single bad market can take most of the account, and the backtest probably doesn't show that market.

5. Enough trades, enough years

Fifty trades over six months proves very little. Look for hundreds of trades across several years, including at least one strongly trending year and one quiet, choppy year. For gold, include a period with a major spike or crash.

6. A live, verified record

A backtest is a simulation. A live account verified on Myfxbook or FXBlue, running for at least six months, shows how the EA copes with real spreads, slippage and server time. Compare it with the backtest for the same period. If live is much worse, find out why before you trust the backtest.

7. Run it yourself, small

Even after all of this, start on a demo account, then on a small live account, for a few months. Keep the risk setting low. If the EA behaves differently from what you expected, you'll find out cheaply.

A quick checklist

1. Real-tick backtest with realistic spreads
2. Out-of-sample results on data the settings weren't tuned on
3. Maximum equity drawdown, not just balance drawdown
4. No martingale, or you understand and accept the tail risk
5. Hundreds of trades across several years and market types
6. A verified live record of six months or more

If you're building your own system rather than buying one, our guide on how to backtest a trading idea covers the traps we've learned to avoid.

Risk note: trading forex carries a high level of risk and may not be suitable for everyone. This article is for education only and is not financial advice.