MT5EAOptimizationGenetic AlgorithmStrategy Tester

A Practical Introduction to Genetic Algorithm Optimization in MT5 EAs

Published: 2026-05-12Read time: about 3 min

⏱ This article reflects information as of its publish date. EA performance figures (PF, DD, annual return) change with live trading and re-validation — check the latest on the EA pages. See the latest EA results →

A Practical Introduction to Genetic Algorithm Optimization in MT5 EAs

When tuning EA parameters, MT5's Strategy Tester offers two optimization modes: exhaustive search (testing every possible parameter combination) and genetic algorithm (GA). When the number of combinations is enormous, genetic algorithm is the standard approach for exploring the space in a realistic amount of time. This article explains the fundamentals of genetic algorithm optimization and the key practical points to keep in mind.

What Is a Genetic Algorithm?

A genetic algorithm (GA) is an optimization technique modeled on biological evolution. The process works roughly as follows:

Generate multiple sets of random parameters ("individuals")
Run a backtest for each individual and score it using an evaluation metric (PF, recovery factor, etc.)
Select the top-scoring individuals as "parents" and produce a next generation through crossover and mutation
Repeat this cycle — with each generation, the parameters converge toward higher-scoring values

Where exhaustive search sweeps through every combination, GA narrows its search to promising regions, cutting computation time by orders of magnitude.

When to Use Each Method

Few parameters with fewer than a few thousand combinations → Exhaustive search for complete coverage
Many parameters with tens of thousands of combinations or more → Genetic algorithm

For example, three parameters each stepped across 10 values yields 1,000 combinations — exhaustive search is perfectly fine. But five parameters each stepped across 20 values produces 3.2 million combinations, at which point genetic algorithm becomes necessary.

Configuring the Strategy Tester

To enable optimization, set the following in the Strategy Tester panel:

Optimization: Select "Fast genetic algorithm"
Result: Choose an evaluation metric such as "Maximum balance," "Custom max," or "Maximum recovery factor"
Inputs tab: Check the parameters you want to optimize and set the start value, end value, and step size

Choosing Your Evaluation Metric

What you define as "good" determines the outcome. Here are the most common metrics and their characteristics:

Maximum balance (profit): Maximizes profit alone. Ignores risk, making overfitting easy to stumble into.
Recovery factor: Net profit ÷ maximum drawdown. Balances return and risk.
Sharpe ratio: Accounts for the stability of returns over time.
Custom max: Freely defined inside the EA code (e.g., PF × trade count).

Using recovery factor or Sharpe ratio is the most practical approach for finding useful parameters while avoiding overfitting.

Running the Optimization

Once optimization starts, the "Optimization Results" tab fills with high-scoring parameter sets, generation by generation. MT5's genetic algorithm automatically runs through hundreds of generations, converging toward top solutions.

Common Pitfalls

Converges too quickly: The parameter search range may be too narrow.
Never converges: The evaluation metric may be inappropriate, or there may be too much noise.
Converges to an obviously strange value: Results with very few trades have not been filtered out.

You can eliminate low-sample noise solutions by baking a filter into your evaluation function — for example, "exclude any result with fewer than 50 trades."

Avoiding Overfitting

The more you optimize, the closer you get to parameters that look perfect on historical data but fail in live markets. Here are practical rules to prevent that.

1. Split Your Data Period

With ten years of historical data, optimize on the first seven years and use the remaining three years as an out-of-sample test. If results deteriorate on the latter period, the strategy has overfit to the earlier one.

2. Compare the Top 10–20 Results

Picking only the single top-scoring result is risky. Check whether the top solutions cluster in a similar region. If they do, adopt a parameter set near the center of that cluster.

3. Check Parameter Sensitivity

Take your candidate parameters and shift them ±10–20%, then verify that performance does not collapse. If it does, that solution is overfit to a narrow "spike" in the optimization landscape.

4. Keep Some Parameters Fixed

Always retain at least a few parameters that are not optimized — values that are naturally justified by the strategy logic. For example, you might optimize the EMA periods while keeping the ATR multiplier fixed. This reduces the risk of overfitting.

Leveraging Parallel Optimization

MT5 supports parallel optimization across multiple CPU cores. Increasing the number of agents under "Tester" → "Settings" reduces computation time.

You can also use the MQL5 Cloud Network to run large-scale optimizations on remote agents worldwide, though this incurs costs. The efficient approach is to first narrow down the parameter range locally, then offload the fine-grained sweep to the cloud if needed.

How We Approach Optimization in Our Free EAs

For the GOLD_EMA_ATR_EA (XAUUSD H1), we deliberately limit what gets optimized:

Short EMA period
Long EMA period
ATR multiplier (for the stop-loss)

All other parameters are fixed at naturally justified values to guard against overfitting. A ten-year backtest produces a PF of 1.30 — that number would climb if we optimized more aggressively, but the probability of it failing in future markets would rise along with it. Choosing modest numbers and running them consistently is the practical answer for live trading.