Forward testing in MT5: what it is and how to set it up
The forward test is the simplest way to know whether an optimization found something real or just fitted the past. MT5 has it built in, but you have to turn it on and know how to read it.
What forward testing is
The in-sample is the data used to choose the parameters. The forward is a later stretch that the optimizer did not use. Since you cannot test on the future, keeping the end of your history aside simulates what would have happened if you had optimized and then traded.
How to turn it on in MT5
In the Strategy Tester (Ctrl+R), settings tab, next to the dates there is the Forward option. You can choose 1/2, 1/3 or 1/4 of the period, or Custom to set the start date yourself. MT5 always takes the final stretch: with a period from January 2020 to January 2024 and 1/4, it optimizes on the first three years and keeps the last one for the forward. It only applies when optimizing.
How much of the period to keep aside
There is no magic number. A quarter or a third is usually a reasonable balance: enough history to optimize and enough forward for it to mean something. What matters is the number of trades: if your EA only makes 10 or 15 in the forward, any result could be chance. If it trades rarely, lengthen the total period rather than shortening the forward.
Which passes are tested
MT5 does not run every pass in the forward, only the best ones of the optimization according to the criterion you chose: 10% in a complete search and 25% in a genetic one. That is why the forward tab has fewer rows than the results tab. Each forward row also carries the result of that same pass in the optimization period, so you can compare them.
How to read the result
It is normal for the forward to be worse than the in-sample: the optimization chose what best fitted the past, and part of that was luck. Good signs: the best passes stay profitable with an acceptable drawdown, and nearby configurations behave similarly. Bad signs: they sink, turn negative, or the ranking is reshuffled completely.
Common mistakes
- Picking the pass with the best forward. The forward stops being an independent test: you have optimized twice.
- Re-running the optimization until the forward looks good. It is the same mistake in more steps: in the end the forward is fitted too.
- A forward that is too short. With few trades it proves nothing, for or against.
- Believing a good forward guarantees the future. It is one more test, not a promise; the cleanest check is a period you used neither to optimize nor to validate.
Next step
Export the results tab and the forward tab to XML and drop them into Orometra. It checks the stable zones it finds in the in-sample against the forward, without mixing or averaging the two periods, and tells you how much evidence backs the configuration. Everything runs in your browser.