What Is FX EA Backtesting? The Right Approach and Key Pitfalls

Before putting real money behind an EA, you need to test it against historical data — that process is called backtesting. A strong backtest result is no guarantee of the same performance in live trading, but running an EA without any backtest is like boarding a plane that skipped its pre-flight check.

This article walks through the correct way to run a backtest using MT5's Strategy Tester and the key pitfalls you need to watch out for.

What Backtesting Can Tell You

A backtest primarily reveals:

• Whether the EA's logic worked in past market conditions
• A rough estimate of maximum drawdown
• The expected trade frequency
• Parameter sensitivity (how much results shift when you tweak a setting)

That said, there is a great deal that backtesting cannot tell you:

• Whether the same performance will hold going forward
• How the EA behaves during a once-in-a-decade shock (2008-level crisis, COVID-scale collapse)
• Slippage caused by a specific broker's order-fill characteristics

The right mindset is not "wins in backtest = wins live." Think of it as a screening filter: an EA that collapsed in historical data has a high probability of collapsing in the future too.

Launching the Strategy Tester

Open the Strategy Tester from the MT5 menu: View → Strategy Tester. The key settings are:

• Expert: Select the EA you want to test
• Symbol: GOLD, EURUSD, etc.
• Timeframe: M15, H1, etc.
• Period: At least 5 years; 10 years is preferred
• Model: "Every tick" is recommended (highest precision)
• Deposit: The starting balance for the test
• Leverage: Match your live trading conditions

"Every Tick" vs. "Every Tick Based on Real Ticks" — What's the Difference?

MT5 offers several modeling modes:

• Every tick based on real ticks: Highest accuracy, but takes the longest to run
• 1-minute OHLC: Fast, with moderate accuracy
• Open prices only: Very fast, but low accuracy — unsuitable for scalping strategies

The shorter your EA's trade duration, the higher the modeling precision you need. A result that looks good on "Open prices only" can deteriorate dramatically when re-run at full tick precision.

Historical Data Quality

The quality of MT5's built-in historical data varies by broker and server. Before running a backtest, verify:

• Modeling quality: 90% or above is desirable
• Data gaps: Older data tends to have more missing bars
• Spread: Use a value close to the real average spread for your broker

Selecting "variable spread" in the backtest means you only test against normal-market spreads — the wider spreads seen during major news releases are not reflected. Since those widenings eat into live profits, another valid approach is to run with a fixed spread set to a realistic worst-case value.

Reading Backtest Results

Here are the key metrics shown after a backtest and what to look for.

Profit Factor (PF)

Total profit divided by total loss. A healthy, realistic range is roughly 1.2 to 1.5. Any EA showing a PF above 3.0 should almost certainly be suspected of curve-fitting.

Maximum Drawdown

The largest peak-to-trough decline in account equity over the test period. Whether you can personally tolerate this figure determines whether the EA is viable for you. Assume that live drawdown will be 1.5 to 2 times worse than the backtest figure.

Number of Trades

A few dozen trades carries no statistical weight. Aim for at least 200 trades, ideally 500 or more to draw meaningful conclusions.

Recovery Factor

Net profit divided by maximum drawdown. A value of 3 or higher suggests the return is proportionate to the risk taken.

Avoiding Curve-Fitting

An EA that has been over-optimized to fit past data will not work in future markets. The basic defenses are:

1. Out-of-Sample (OOS) Validation

Split your data: for example, optimize on 2015–2022 and test on 2023–2025. If the out-of-sample period delivers comparable results, the risk of over-optimization is much lower.

2. Check Parameter Sensitivity

If shifting the optimal parameter by ±10–20% causes a dramatic collapse in results, that is a warning sign. Look for parameters that sit in a flat plateau — a region where small changes still produce stable outcomes.

3. Test Across Multiple Symbols and Timeframes

An EA that only works on one specific symbol and timeframe may simply be over-fitted to those conditions.

Backtesting the EAs on This Site

The GOLD_EMA_ATR_EA (XAUUSD H1) 10-year backtest results are as follows:

• Test period: 2015–2025 (10 years)
• Profit Factor: 1.30
• Win rate: 49%
• Maximum drawdown: 5.88%
• Annual return: 1.7%
• Modeling quality: 99.9% (every tick, variable spread)

The numbers are modest, but the priority here is stability — surviving every market phase over 10 years without a collapse. The full report is available on the download page.

The Importance of Forward Testing

Once a backtest shows solid results, run a 3 to 6 month forward test on a demo account or micro account. Testing against real-time fills, real slippage, and actual news-event behavior lets you catch bugs and unexpected behavior before committing live capital.

Free EA Download

GOLD_EMA_ATR_EA is available for free download, complete with the full 10-year backtest report.

Download the free EA

Recommended Brokers

To run backtests and forward tests in the same environment, this site recommends verified brokers.

View the recommended broker list