Before you trust any robot with real money, you have to backtest it. A backtest replays a MetaTrader 5 Expert Advisor over historical price data so you can see how it would have behaved — its profit, its drawdowns, and how often it traded. This step-by-step guide shows you how to backtest an MT5 EA properly in 2026, which metrics actually matter, and the mistakes that fool beginners into trusting a robot that later fails live.
This is the verification step we recommend in every guide, including the best MT5 EAs of 2026 and the safe-EA buyer's guide.
Step-by-step: backtest an EA in the Strategy Tester
- Open the Strategy Tester — press
Ctrl+Ror View → Strategy Tester. - Select the EA from the Expert dropdown.
- Set the symbol and timeframe — e.g.
XAUUSDandM15. - Choose a date range — use several years to capture different conditions (trending, ranging, news-heavy).
- Set modelling to "Every tick based on real ticks" when your broker provides real tick history — it's the most accurate model.
- Set a realistic deposit, leverage and currency, and configure the EA inputs (lot/risk, etc.).
- Disable optimisation for a single, honest run.
- Run it, then open the Backtest and Graph tabs to read the report.
The metrics that actually matter
| Metric | What it tells you | Rough guide |
|---|---|---|
| Net profit | Total gain over the period | Positive, but read with drawdown |
| Profit factor | Gross profit ÷ gross loss | >1.3 decent, >1.5 good |
| Max drawdown % | Worst peak-to-trough drop | Lower is better; can you stomach it? |
| Total trades | Sample size | 100+ for any statistical meaning |
| Expected payoff | Avg profit per trade | Positive after costs |
| Recovery factor | Net profit ÷ max drawdown | Higher = more efficient |
The trap: a huge net profit with a 70% drawdown and only 25 trades is not a good EA — it's a small, lucky (or curve-fit) sample. Always read profit against drawdown and trade count.
Modelling & data quality matter
Garbage data produces garbage results. For scalping EAs especially, use real tick data and set a realistic spread — testing on an unrealistically tight spread will overstate results. If you can only use generated ticks, treat the numbers as a rough sketch, not proof.
5 common backtesting mistakes
- Curve fitting — optimising inputs until the past looks perfect. It won't repeat. Test on data the EA wasn't tuned on.
- Ignoring spread & commission — add your broker's real costs; they erode scalping edges fast.
- Too short a period — a few months can't show how an EA handles different regimes.
- Lowest-quality modelling — fine for a glance, misleading for decisions.
- Trusting the backtest alone — see below.
Backtest, then forward-test on demo
A backtest is necessary but never sufficient. Real fills, slippage, latency and live spreads differ from history. After a solid multi-year backtest, forward-test on a demo account on your own broker for several weeks, then go live with small size under the 1% risk rule. This is the same honest sequence we recommend instead of trusting advertised win rates — see the debate in EA vs manual trading.
Backtesting Gold Scalpers EA
One practical advantage of Gold Scalpers EA is that the EMA indicator it uses is bundled inside the EA — so it runs in the Strategy Tester and on any PC with no extra files or "indicator not found" errors. Drop it on XAUUSD, set M15, pick a multi-year range with real ticks, and you can verify exactly how the 9/21 + 50-EMA logic and the swing stop would have behaved — then demo-test before going live. That's the whole point: judge it on your own data, not our word.
Frequently asked questions
How do I backtest an EA in MetaTrader 5?
Open the Strategy Tester (Ctrl+R), pick the EA, symbol and timeframe, set a multi-year range with real-tick modelling, enter a realistic deposit and inputs, then run and review the report.
What is a good profit factor?
Above 1.3 is decent, above 1.5 is good — but only with 100+ trades over several years and realistic costs.
Is backtesting enough before going live?
No. Always forward-test on demo afterward, then start live small.
What modelling quality should I use?
"Every tick based on real ticks" when available; avoid drawing strong conclusions from the lowest-quality model.
Why do backtest and live results differ?
Spread/commission, slippage, latency, broker execution, and curve fitting. Forward-testing bridges the gap.