Testing guide
How to properly test MT5 Expert Advisors
Use the Strategy Tester to challenge the EA — not to invent a perfect curve. This guide covers data quality, costs, forward checks, and how to read the report without self-deception.
Evidence over comfort
A beautiful backtest with poor modelling quality, zero costs, or no forward window is marketing — not evidence. Prefer conservative assumptions.
Step-by-step guide
Data · Costs · Forward · Report
- 1
Define what you are testing
Write the hypothesis: symbol, timeframe, risk policy, and the go/no-go criteria before you open the tester.
- 2
Install the EA and open Strategy Tester
Confirm the build is in Experts, then open View → Strategy Tester (Ctrl+R).
- 3
Download real tick history
Use high-quality ticks from your broker or MetaQuotes. Incomplete history is the fastest path to false confidence.
- 4
Set modelling to every tick / real ticks
Prefer real ticks when available. Coarse modes are only for smoke tests, not validation.
- 5
Include spread and commission
Model realistic costs. Zero-spread, zero-commission runs systematically overstate expectancy.
- 6
Choose a multi-regime date range
Include calm and volatile periods. A short cherry-picked window is not proof.
- 7
Run a baseline with documented parameters
Start from Finversa defaults or a named `.set`. Save the report before any optimization.
- 8
Read drawdown before net profit
Max drawdown, consecutive losses, and recovery time matter more for go-live than total profit.
- 9
Run a forward / hold-out check
Reserve unseen dates. If performance collapses out-of-sample, the in-sample curve was fitting noise.
- 10
Optional walk-forward split
For optimization, rotate train/test windows. Full-history optimization without hold-out is a common failure mode.
- 11
Compare to demo Track Record
Treat tester results as a hypothesis. Confirm behavior on demo with real spreads and fills before sizing up.
- 12
Document the decision
Record build, dates, costs, metrics, and the go/no-go call. Continuity beats memory when you revisit later. Open backtesting guide