How to Evaluate a MetaTrader 5 Expert Advisor Before You Risk Real Money
Every week, new Expert Advisors appear with equity curves that rise in a straight line. Most of them will not survive their first year on a live account. The problem is rarely the code. It is the evaluation: buyers and developers alike look at the wrong numbers, in the wrong order.
This post is the checklist I use on my own EAs and on every EA I am asked to review. None of it requires special software, only the Strategy Tester and some discipline.
1. Start with the number of trades, not the profit
A backtest with 80 trades says almost nothing. With so few observations, a strategy with zero edge can easily produce a profit factor above 1.5 by chance. Before reading any other metric, check that the sample is large enough: several hundred trades, spread over several years and several market regimes (trending, ranging, high and low volatility).
If the EA trades rarely by design, extend the test period or test it on more symbols. If it only works on one symbol and one period, that is information too.
2. Read expectancy and profit factor together
Expectancy tells you what the strategy earns on average per trade, after losses. Profit factor tells you how gross profit compares to gross loss. Look at both:
- A high win rate with a profit factor near 1.1 usually means small wins and occasional large losses. One bad week can erase months.
- A low win rate (35 to 45 percent) with a profit factor around 1.4 to 1.8 is typical of trend and breakout systems. It is psychologically harder, but often more robust.
A win rate on its own is the least informative number in the report.
3. Treat maximum drawdown as a minimum
The maximum drawdown in a backtest is the worst path that happened, not the worst path that can happen. Shuffle the order of the trades (a simple Monte Carlo resampling of the trade list) and you will usually find drawdowns 1.5 to 2 times deeper than the historical one.
Size your risk on that resampled figure. If you trade a prop firm account with a daily loss limit, check the worst single day as well, not only the overall drawdown.
4. Check the cost model
A large share of "profitable" EAs are profitable only with zero or fixed spread. Re-run the test with:
- Every tick based on real ticks, when the broker history allows it.
- Realistic commission per lot.
- A spread that widens around rollover and news.
If the results collapse when spread doubles, the edge is thinner than the costs you will pay live.
5. Look for fragile parameters
Optimize one parameter at a time and look at the neighbourhood of the chosen value. A robust parameter sits on a plateau: values around it give similar results. A fragile one sits on a peak: move it by one step and the profit factor drops below 1. Peaks are a symptom of curve fitting.
6. Separate in-sample and out-of-sample
Keep the most recent period (for example the last 20 to 30 percent of your data) completely outside the optimization. Run it once, at the end. If performance degrades sharply, the optimization learned noise. Walk-forward analysis is the more rigorous version of the same idea.
7. Forward test on demo, then on a small live account
A backtest cannot model slippage, requotes, server outages or your broker's execution. A few weeks on demo, then a small live allocation, is the only way to measure the gap between theory and reality.
Summary checklist
- Enough trades, several years, several regimes.
- Expectancy and profit factor, read together.
- Drawdown stress-tested by resampling.
- Realistic spreads and commissions.
- Parameters on plateaus, not peaks.
- Clean out-of-sample period.
- Demo, then small live.
No EA passes this list perfectly, including mine. The point is not perfection, it is knowing exactly which risks you are accepting.
Disclosure: I am the developer of Titan Breakout, an H1 breakout Expert Advisor for MT5, published by Pipmaster France. It can be downloaded for free for backtesting and demo testing, so you can apply this checklist to it yourself: pipmaster.fr.


