Backtesting an EA: Why Past Results Don't Guarantee Future Performance

A backtest is the foundation of any serious EA evaluation, but it is also the most misunderstood step. Many buyers see an impressive backtest and treat it as a forecast. It is not. A backtest is a description of how a specific ruleset would have behaved on a specific dataset under specific assumptions. That is useful. It is not the same as a promise. This article explains why backtests can mislead, how to spot the ones you can actually trust, and where the gap between backtest and live sits, so you can approach EA evidence with the scepticism it deserves.

What a Backtest Actually Does

A backtest replays historical price data through an EA’s decision logic and records the resulting trades. In MetaTrader, this is done through the Strategy Tester. The official MT5 Strategy Tester documentation explains that it is designed to test and optimise strategies before live trading, using multi-currency, multi-threaded historical simulations.

A backtest tells you what the rules would have done. It does not tell you what future markets will do. That single distinction separates people who use backtests well from people who lose money to them.

Why Past Results Cannot Guarantee Future Ones

Financial markets are not stationary. Volatility regimes change, correlations shift, central bank policy moves, and liquidity migrates. An EA that thrived during one regime can quietly stop working when the regime ends. Common reasons a backtest fails to predict live results:

• Curve fitting. Parameters were over-optimised to fit historical noise rather than a real edge.
• Data quality. Ticks in the tester are modelled, not identical to what your broker will send.
• Spread and commission assumptions. Default tester spreads are often lower than live spreads.
• Slippage. Testers rarely model realistic slippage on stops and news events.
• Survivorship in the sample. The developer only ships EAs whose sample results looked good, which is itself a selection bias.
• Server-time and DST shifts. These can move signal bars in ways the tester does not replicate live.

The Difference Between a Good Backtest and a Convincing One

A convincing backtest is not the same as a good one. The features that make a backtest genuinely useful are boring rather than exciting.

Trait	Convincing (but weak)	Genuinely useful
Sample length	6 months	5+ years
Instruments tested	1 pair	Multiple correlated and uncorrelated
Modelling mode	Open prices only	Every tick or real ticks
Spread setting	Fixed 0.1 pip	Realistic variable spread
Optimisation approach	Grid-searched over the whole sample	Walk-forward, out-of-sample validation
Equity curve shape	Straight line	Realistic drawdown periods
Trade count	20	Hundreds or thousands
Commission modelling	Ignored	Applied at broker-realistic rate

A statistically meaningful sample usually needs at least a few hundred trades across different market conditions. Twenty trades over three months tells you almost nothing.

Walk-Forward and Out-of-Sample Testing

One of the most important techniques is walk-forward analysis: optimise the EA on a slice of history, then run it on a later slice that was never seen during optimisation, then step forward. If the EA still works out-of-sample, the edge is more likely real. If it works only in-sample, it is likely curve-fitted.

A common shortcut is to split the data 70/30: optimise on the first 70%, then check performance on the untouched 30%. Any strategy that fails this basic split should not go live. Serious developers automate this with multiple rolling windows to test consistency across different periods.

Live vs Backtest: What to Expect

Even a well-conducted backtest usually overstates live performance. A reasonable working assumption is:

• Live returns will be lower than backtest.
• Live drawdown will be deeper than backtest.
• Live Sharpe ratio will be lower than backtest.
• Live win rate may be similar, but average win size will typically shrink and average loss size will grow.
• Live trade count may be higher or lower than backtest depending on execution filters.

This does not mean backtests are useless. It means they are a floor for scepticism, not a ceiling for expectations. Our editorial approach to EA evaluation sets out the specific tests we run before we accept a backtest as evidence of anything.

Forward Testing Bridges the Gap

A well-structured EA evaluation ends with forward testing on a demo or micro-live account for at least three to six months. Forward testing exposes the EA to real spreads, real slippage, and real execution while the backtest assumptions are still fresh. Meaningful divergence between backtest and forward test is a strong signal that the backtest was overfitted.

Practical Backtest Checklist

• Sample of at least several years covering different regimes.
• Every tick modelling or real ticks.
• Realistic variable spread, ideally from your target broker.
• Commission modelled where applicable.
• Slippage simulated on stops.
• Out-of-sample validation.
• Multiple instruments where the strategy claims to be robust.
• Consistent risk per trade throughout.
• Forward test comparison against the backtest.

If any of these are missing, treat the results as a marketing artefact rather than evidence.

FAQ

Is a backtest useless then?
No. A backtest is essential. It rules out strategies that fail on history. But passing a backtest does not certify success.

How long a backtest is enough?
At minimum enough to include multiple volatility regimes. Several years is a reasonable working baseline.

Can I trust a developer’s own backtest?
Only if you can reproduce it yourself with your own data.

What is curve fitting?
Curve fitting is over-optimising parameters until they perfectly match past data at the cost of losing generality.

Should I run my own backtest?
Yes. It is the single most important step before deploying any EA on real money.

Does using real tick data eliminate backtest bias?
It reduces one source of bias significantly, but not curve fitting, survivorship bias, or execution modelling gaps.

Trading foreign exchange and CFDs on margin carries a high level of risk and may not be suitable for all investors. This article is for educational purposes only and does not constitute financial advice.

Evaluate EAs Against Real Backtest Standards

We hold EAs to a rigorous backtest and out-of-sample standard before we consider recommending them. See our current vetted Expert Advisors editorial position to understand how the tests are applied.