How to Backtest an Expert Advisor in MetaTrader 5 (Strategy Tester deep dive)

Instructions étape par étape

1. Étape 1 : Open the Strategy Tester

   View → Strategy Tester (Ctrl+R). The tester panel appears at the bottom of MT5 with tabs Settings, Inputs, and Results.

   The Strategy Tester is a separate engine from the live terminal — it has its own copy of the EA's compiled code, its own historical tick stream, and runs in a sandbox with no network access. This is by design: a backtest must be deterministic and reproducible.

   First-time use requires MT5 to download historical data for the symbol + timeframe combination you pick. The first backtest of any pair triggers a download of M1 ticks for the entire requested date range, which can take 5–20 minutes depending on the broker and date range. Subsequent runs reuse the cached data and start instantly.

2. Étape 2 : Configure the Settings tab

   On the Settings tab, fill in: Expert (pick your EA from the dropdown), Symbol (pick from your broker's symbols), Period (the timeframe; usually match the live attach timeframe), Modelling (see next step), Date (From/To), Forward (split for walk-forward testing), Deposit (the initial balance), Leverage (the broker's actual leverage on your account type), Currency (account currency), Optimization (off for a single backtest).

   Date range matters a lot. For a credible backtest, use a minimum 5-year window that includes both trending and ranging regimes — typically 2018-01-01 to today covers the 2020 COVID volatility spike, the 2022 inflation regime change, and the 2024+ rate-cut cycle. Anything shorter risks fitting to one regime.

   Deposit should match your live account — 0.1 lot trades produce very different psychological pressure on $1000 vs $100,000 even when the absolute P&L is the same; lot-percent risk models will scale accordingly.

3. Étape 3 : Pick the right Modelling mode

   Modelling determines how MT5 reconstructs each price tick during the backtest. There are four modes:

   'Every tick' — uses M1 OHLC to interpolate every tick. Fast, but ticks are synthetic. Profit factor on scalping EAs is often 1.5–2× too high. Avoid for scalping.

   'Every tick based on real ticks' — uses the broker's actual recorded tick history. Slow (10× longer than synthetic) but produces backtests that closely match live performance. Use for any EA on M1/M5 timeframe or any scalping strategy.

   'OHLC 1 minute' — uses only the open, high, low, close of each M1 bar. Very fast but the order of high vs low within the bar is undefined, which breaks any EA that uses intra-bar logic. Useful only for daily-timeframe EAs.

   'Open prices only' — fastest but unrealistic. Use only for sanity-checking initialisation logic, never for performance evaluation.

   Default to 'Every tick based on real ticks' unless you are time-constrained or running a daily-timeframe EA.

4. Étape 4 : Load the vendor preset on the Inputs tab

   Switch to the Inputs tab. Click Load and pick your vendor .set file from MQL5/Presets/. Verify the loaded values are appropriate for the backtest scenario.

   For a fair backtest you want the EA configured exactly as it would run live. That means: same lot size, same risk percent, same time filters, same magic number. The only setting you should override for the backtest is the trading-window date if the EA hard-codes 'do not trade before 2024' or similar.

   If the EA has a 'Backtest Mode' or 'Optimization Mode' input that disables certain checks, leave it off. Those modes exist for fitting and produce unrealistically good backtest results.

5. Étape 5 : Set realistic spread and commission

   Strategy Tester defaults to 'Current spread' which uses the live tick spread at the moment you start the backtest. This is unrealistic — spreads widen during news, during low-liquidity hours (Sydney open), and during fast moves.

   For a pessimistic backtest, override the spread to a fixed point value 50–100% above the typical spread. For EURUSD on ECN, typical is 0.3 pips so use 0.6 pips (6 points). For XAUUSD, typical is 20 points so use 40 points. The resulting backtest is conservatively pessimistic.

   Commission is set via Tools → Options → Charts → 'Commission, $ per 1.0 lot per side' — but this only affects the live terminal display. For Strategy Tester commission, the broker's tick data file usually includes commission metadata. If you need to override, edit the symbol specification in `Symbols → Properties → Trade → Commission`.

6. Étape 6 : Run the backtest

   Click Start. The Strategy Tester begins streaming historical ticks through the EA. Progress is shown as a bar at the bottom; the Equity curve updates in real-time on the Graph tab.

   A 5-year EURUSD M5 backtest with Every-Real-Ticks takes about 15 minutes on a modern PC. XAUUSD M1 backtests are slower because the tick density is 5–10× higher than EURUSD. If your machine is single-core (e.g. Mac via Wine), expect 3–5× longer durations.

   Watch the Journal tab during the run for errors. Common runtime errors are 'Cannot load history' (data still downloading; pause the backtest, wait, restart) and 'Symbol not found' (symbol name mismatch). Successful runs complete with 'Tester result: ...' in the log.

7. Étape 7 : Interpret the Results tab

   When the backtest completes, the Results tab shows summary metrics. The key fields to read, in order of importance:

   Profit Factor = gross_profit / gross_loss. 1.5+ is acceptable, 2.0+ is good, 3.0+ is suspicious (probably overfit). Below 1.3, the EA loses money after spreads and commissions.

   Maximum Drawdown — the biggest peak-to-trough equity drop. Should not exceed 30% for any retail EA; 50%+ means the EA blows up an average account during a typical bad year.

   Sharpe Ratio (visible in MT5 build 4000+) — risk-adjusted return. Above 1.0 is acceptable; 2.0+ is good. The MT5 calculation uses daily returns; backtests under 1 year may give misleading values.

   Recovery Factor = net_profit / max_drawdown. Above 5 is good — means the EA recovers from its worst drawdown 5× over. Below 2 is fragile.

   Total Trades — backtest credibility scales with trade count. Below 100 trades, the metrics are noisy. Aim for 500+ trades over a 5-year window.

8. Étape 8 : Save the backtest report

   Right-click anywhere in the Results tab → 'Save as Report...'. MT5 generates an HTML report with the full equity curve, monthly returns, drawdown profile, trade list, and all summary metrics. Save the report somewhere persistent — these are the documents you compare future backtests against.

   For systematic tuning, also save the .set file via the Inputs tab → Save button, named with the backtest date and key metric, e.g. `MyEA-Aggressive-PF1.82-DD18-2026-05-17.set`. This way you have a paper trail of which preset produced which backtest.

   Before acting on a backtest, run the same EA in walk-forward mode (Settings → Forward, split 50/50). Walk-forward results meaningfully diverging from in-sample results are the single most important indicator that the in-sample numbers are overfit.

Erreurs courantes à éviter

• ✗ Using 'Every tick' (synthetic) modelling for scalping EAsSolution: Always use 'Every tick based on real ticks' for sub-H1 timeframes. Synthetic ticks miss the spread widening that kills scalpers in live.
• ✗ Backtesting on 1 year of data and treating the result as predictiveSolution: 5 years minimum. Single-year backtests have too few regime changes to be statistically meaningful.
• ✗ Leaving spread at 'Current' instead of a pessimistic fixed valueSolution: Pad spread 50–100% above typical. Backtests with tight spreads look great but live execution loses 0.5+ pips per round trip.
• ✗ Ignoring commission entirelySolution: ECN accounts charge $3–7 per round-turn lot. For scalpers this is 30–50% of net profit; not modelling it is the biggest single source of optimistic backtest bias.
• ✗ Optimizing on the same period you backtest on (in-sample fitting)Solution: Use walk-forward: optimize on 70% of the period, validate on the remaining 30%. Discard any preset that doesn't survive the out-of-sample test.
• ✗ Backtesting only on a single symbol when the EA trades a basketSolution: MT5 backtests one symbol per run. For basket EAs, backtest each symbol separately and aggregate the results in a spreadsheet.
• ✗ Trusting Net Profit as the primary metricSolution: Net Profit can be inflated by an outsized lot size on a fluke trade. Use Profit Factor and Recovery Factor as primary; Net Profit as sanity check.

Questions fréquemment posées