How to Backtest an Expert Advisor on MT5 (Without Getting Fooled by Fake Results)
A backtest can lie. Overfitting, tick quality, missing slippage, the traps are everywhere. Here's how to read an MT5 backtest correctly, and what separates a solid validation from a fabricated equity curve.
An impressive backtest doesn't mean much if you don't understand how it was built. That's the number one problem in the Expert Advisor space: dozens of vendors display near-perfect equity curves that collapse within the first month of live trading.
This guide explains how to read an MT5 backtest correctly, and how to tell the difference between a solid result and a manufactured one.
Why Backtests Can Lie
Before looking at the numbers, you need to understand the structural biases of the MT5 Strategy Tester.
The tick quality problem
MT5 offers three modeling modes:
- Every tick: Generates synthetic ticks from OHLC bars. Fast, but ticks between open and close are interpolated.
- Every tick based on real ticks: Uses actual ticks downloaded from the broker. Slow, but the only mode that reflects reality.
- Open prices only: Only useful for strategies that operate exclusively at candle open.
The majority of published backtests are run in standard "Every tick" mode. The problem: MT5 itself reports that real tick quality in this mode is often below 10%. The remaining 90% is interpolation.
What this means in practice: the ultra-precise entries and exits that make the backtest shine never actually existed. Spread, slippage, and gaps are not faithfully reproduced.
Overfitting (curve fitting)
An EA can be optimized on historical data until it produces a perfect equity curve. This is technically trivial. But that EA has zero robustness, it memorized the past rather than learning the logic of the market.
Warning signs:
- Win rate above 85% with profit factor above 5 over long periods
- Very precise parameters (e.g., TP = 127 points exactly)
- Performance drops sharply when you shift the test period by a few weeks
- No economic logic behind the strategy
Missing commissions and swap
By default, the Strategy Tester doesn't charge commissions or swap fees. On a frequently trading EA, the impact can be massive.
Example: an EA with 1,000 trades over 3 years at €3 commission per round-trip = €3,000 in unaccounted fees. On a reported gain of €10,000, that's a 30% discrepancy.
How to Read an MT5 Backtest Properly
The metrics that actually matter
Profit Factor: ratio of gross profit to gross loss. Below 1.5, the system doesn't adequately offset losses. Above 2.5, it's solid, provided the trade count is meaningful.
Max Drawdown: the maximum peak-to-valley loss. Look at the equity drawdown percentage, not just the absolute value. A 15% DD is manageable. 40% is a different story.
Recovery Factor: net profit divided by max drawdown. Measures how well the system recovers from losses. A recovery factor above 5 is a good sign.
Trade count: below 300 trades, the statistics are not reliable. A system with 30 trades and a 90% win rate proves nothing statistically.
Z-Score: measures whether winning and losing streaks are random or correlated. A negative Z-Score (e.g., -2.4) with high probability (> 95%) indicates that losses and wins don't cluster, a good property for a mechanical system.
LR Correlation: measures the regularity of growth. Correlation close to 1 = smooth, predictable equity curve. Correlation at 0.5 = erratic equity even if the final result is positive.
The credible backtest checklist
| Criterion | Minimum acceptable | Ideal |
|---|---|---|
| Modeling mode | Every tick (real ticks) | Every tick based on real ticks |
| Test period length | 2 years | 3+ years |
| Trade count | 300+ | 1,000+ |
| Profit Factor | > 1.5 | > 2.0 |
| Max equity DD | < 25% | < 15% |
| Slippage simulated | Yes | Yes |
| Commissions included | Yes | Yes |
What to demand before buying an EA
- The backtest configuration file: mbol, timeframe, exact dates, parameters used. Without this, you can't reproduce it.
- Confirmation of "real ticks" mode: screenshot of the Strategy Tester settings is enough.
- Commissions applied: the vendor can't tell you what spread and commission were used, be suspicious.
- An out-of-sample backtest: e test period should include data that wasn't used for optimization. Ask if any portion of the data was reserved for validation.
What BreakEdge Gold Pro Does
The BreakEdge Gold Pro backtest was run in "Every tick based on real ticks" mode on MetaTrader 5, covering October 2022 – April 2026 (3.5 years), with slippage simulation enabled. The strategy is a pure range breakout, no parameters were optimized on this period to inflate the numbers.
Results are publicly viewable, and since April 2026, the EA has been running on a live Vantage demo account under real conditions, independently verified on Myfxbook.
Systematic gold trading: discover BreakEdge Gold Pro and its verifiable results.