How long should I forward test an EA?

Until it has produced enough trades for the result to be more than luck, and long enough to include a period the strategy dislikes. For an active strategy that is often four to eight weeks; for one that trades a few times a month, several months. Counting trades is more useful than counting weeks.

Should I forward test on demo or on a small live account?

Both, in that order. A demo proves the EA runs, the settings are right and nothing is rejected. Only a live account, even a tiny one, shows you real spread, slippage and fills, which is precisely what a demo cannot reproduce.

How many trades make a result meaningful?

Under about 30 trades you are reading noise. Around 100 the average starts to mean something. For a strategy that wins often and loses rarely and largely, even 100 winning trades say little, because the event that matters has not happened yet.

What if the forward test makes less than the backtest?

That is the normal outcome and not by itself a problem. Costs are real now. What matters is whether the shape matches: similar win rate, similar drawdown pattern, similar frequency. A forward test that is quieter but recognisably the same strategy has passed.

Forward Testing an EA: How Long Before You Trust It

A backtest tells you what a strategy would have done on someone else's price history. A forward test tells you what it does on your broker, with your spread, your fills and your latency. The uncomfortable part is how long it takes to mean anything, and the answer depends less on the calendar than on how the strategy loses.

Last updated 6 October 2026

What a forward test is for

It answers questions a backtest structurally cannot:

What is the real spread on this account, at the hours this EA trades?
How much slippage is there between the price requested and the price filled?
Does the broker reject anything — stops too close, symbol restrictions, session gaps?
Does the EA survive a weekend, a restart, a VPS reboot, a MetaTrader update?
Does the news filter behave, given that it cannot run in the tester at all?

None of those are strategy questions. They are plumbing questions, and plumbing is what breaks first.

Two stages, in order

Stage one: demo, one to two weeks

The goal here is not profit, it is absence of surprises. Attach the EA with the settings you intend to use and watch for: orders placed and closed cleanly, no rejections in the Journal, the panel or log reporting sensible state, and the EA still running after a restart.

If the EA has time-based settings, this is also where you confirm they line up with the broker's clock rather than yours — see MetaTrader server time.

Two weeks of clean demo means the installation is right. It means nothing about whether the strategy works.

Stage two: live, small, for as long as it takes

Fund the smallest account the broker allows and run the EA at its minimum lot. Yes, the returns are meaningless at that size. That is not what you are measuring.

You are measuring the gap between requested and filled prices, the spread you actually pay, and whether anything behaves differently now that the orders are real. Those numbers scale; the profit does not need to.

How long, concretely

Count trades rather than days, and adjust for how the strategy makes its money.

Strategy shape	Read the result at	Why
Many small trades, tight stops	100+ trades	The average converges quickly; costs dominate.
A few trades a week, wide stops	40+ trades, 3 months	Fewer samples, each one matters more.
Wins often, loses rarely and largely	Until it has had a bad period	Winning months are not the test. The tail is.

That third row is the one that applies to most grid and martingale EAs, ours included, and it is why "six profitable months" is weaker evidence than it sounds. The strategy is designed to produce exactly that until it does not. What you are waiting for is the first drawdown of any size, so you can see how it recovers and whether the account-level limits behaved.

What to compare at the end

Put the forward test beside the backtest over the same period if you can, and compare four things:

Win rate. Should be close. A large gap means fills are worse than modelled.
Average win and average loss. Both shrink live; the ratio between them should hold.
Trade frequency. Far fewer trades live usually means a filter is blocking entries — spread, session or news.
Maximum drawdown, as a percentage. The number you will live with.

A forward test that earns less but keeps the shape has passed. One where the win rate collapses or the frequency halves is telling you the backtest was measuring a market you cannot trade.

Mistakes that waste the whole test

Changing settings midway. The clock restarts. If you must change something, note the date and treat what came before as a separate test.

Intervening by hand. Closing a trade because you did not like it makes the result yours, not the EA's.

Stopping after the first loss. Losses are part of the sample. Stopping at the first one guarantees you only ever see the start of the distribution.

Testing at a size you cannot sustain. If the position size is large enough to worry you, you will intervene, and then you have not tested the EA.

When to stop the test and walk away

Three signals that do not need more time:

Rejections or errors that never appeared on demo and have no explanation.
A drawdown deeper than the whole backtest showed, early in the test.
Behaviour that does not match the description — entries at times it should not trade, sizes that are not what the settings say.

The first two are information. The third is a reason to distrust everything else you were told, and spotting a forex EA scam covers what to check next.