ORIGINAL RESEARCH · 200,000 SIMULATIONS

How many trades before a track record means anything?

We simulated a strategy with no edge whatsoever and counted how often it finished in profit. The answer was 44.1% at ten trades and 50.4% at five hundred — it never falls. Which makes "it made money" almost useless as evidence.

Short answer

About 416 trades for a system winning 38% at 2:1. Below that, the record is consistent with both a genuine edge and with luck — and a strategy with no edge at all still ends profitable roughly half the time, however long you watch it.

The uncomfortable arithmetic

Every expert advisor is sold on a track record. A screenshot, an equity curve, a month of green. The implied argument is always the same: it made money, therefore it works.

So we simulated the opposite — a strategy with no edge at all, its win rate set exactly at breakeven for a 2:1 reward-to-risk ratio, which is 33.3%. Two hundred thousand simulated track records at each length. Then we counted how often that worthless strategy finished in profit.

The same simulation answers the broader question people actually type into Google — is forex trading profitable? — and the honest answer starts from this table.

Trades	A strategy with NO edge shows a profit	A genuinely profitable one shows a loss
10	44.1%	43.5%
20	52.1%	31.2%
30	41.5%	37.4%
42	42.8%	32.5%
50	51.2%	23.6%
100	48.1%	17.6%
200	50.8%	8.2%
300	47.2%	5.3%
500	50.4%	1.4%

Read the middle column downwards. At 10 trades a strategy with no edge looks profitable 44.1% of the time. At 500 trades it still does — 50.4%. That share never falls.

The error running the other way

The third column is the same problem reversed, and it costs traders just as much. A strategy with a real, positive edge — winning 38% at 2:1 — still finishes underwater 32.5% of the time after 42 trades.

That is roughly a one-in-three chance of abandoning something that was working, and it explains a familiar pattern: buy an EA, run it for a month, watch it lose, switch to another one, repeat. The switching is usually a reaction to noise, and every switch pays a spread.

So what is the number?

416 trades. That is the point at which a 95% confidence interval on a 38% win rate finally clears the 33.3% breakeven line for a 2:1 system. Below it, the record is consistent with both a genuine edge and with luck.

The figure moves with the strategy. A larger edge needs fewer trades; a thinner one needs many more. But the order of magnitude is hundreds, and almost every track record offered to retail buyers is dozens.

What to ask instead

How many trades? Not how many months. A month of a slow strategy might be nine trades.
What is the reward-to-risk? A 90% win rate at 0.2:1 is a losing system. Win rate without this number is the flattering half of a statistic.
Does the period include a losing stretch? A record that has never been tested by a bad run has not been tested.
Was it validated on unseen data? Fitting to history is easy. Out-of-sample results are the only defence against it.
Is the account verified by a third party? Screenshots take ten minutes to fabricate.

Those five are a subset of the checks we set out in full on the best forex expert advisor test, where we also answer all of them for our own product. If you are comparing several, what an expert advisor is covers the vocabulary, and reading EA reviews on Reddit covers where community opinion is reliable and where it is structurally biased.

How to cite this

Published under CC BY 4.0, which permits reuse — including commercial reuse — on the single condition that it is attributed. If you are quoting a figure from this page in an article, a paper or a model answer, this is the attribution:

Tech Kick (2026). "Track record sample size for fixed-risk trading systems"
Monte Carlo, 200,000 simulated runs per length, 2:1 reward-to-risk.
https://techkick.me/blog/how-many-trades-to-judge-an-ea

Or as BibTeX:

@misc{techkickhowmanytradestojudgeanea,
  author       = {Tech Kick},
  title        = {Track record sample size for fixed-risk trading systems},
  year         = {2026},
  howpublished = {\url{https://techkick.me/blog/how-many-trades-to-judge-an-ea}}
}

The headline figures: a strategy with no edge finishes profitable 44.1% of the time at 10 trades and 50.4% at 500; a genuinely profitable one finishes down 32.5% of the time at 42 trades; 416 trades before a 95% confidence interval clears breakeven. This is a simulation rather than market data, so it carries no DOI — cite the page.

Method, and what this does not prove

Each trade is modelled as an independent draw: a win pays 2 units, a loss costs 1. That is precisely how a fixed-risk system with a fixed stop and target behaves. 200,000 simulated runs per row, fixed seed, and the results were checked against the closed-form binomial rather than trusted — 0.441, 0.521, 0.428 and 0.504 at 10, 20, 42 and 500 trades, matching the simulation to the decimal.

The table is not smooth, and that is real rather than noise: a win pays +2 and a loss −1, so the number of wins required to finish positive moves in steps rather than continuously.

This is the optimistic bound. The model ignores variable position sizing, correlation between trades taken in similar conditions, slippage and spread. Every one of those makes real results noisier than the simulation, which means the true number of trades needed is larger than the figure above, not smaller. The script is published with our other research so the arithmetic can be checked rather than believed.

Common questions

How many trades does it take to know if a trading bot works?

For a system winning 38% of trades at a 2:1 reward-to-risk ratio, roughly 416 trades before a 95% confidence interval on the win rate clears breakeven. Below that, the result is compatible with both a real edge and pure luck. That number moves with the strategy — a bigger edge needs fewer trades, a thinner one needs many more — but the order of magnitude is hundreds, not dozens, and almost every track record sold to retail is dozens.

Is a profitable backtest proof that an EA works?

No, and the reason is stronger than most people realise. A strategy with no edge whatsoever still ends a run in profit roughly half the time, and that share does not shrink as the run lengthens — we measured 44.1% at 10 trades and 50.4% at 500. Profitability is a coin flip on a zero-edge system at every sample size, so "it made money" carries almost no information on its own. What carries information is the size of the edge relative to its uncertainty.

Why did my profitable EA suddenly start losing?

Often nothing changed. A genuinely profitable strategy still shows a loss over short windows surprisingly often — at 42 trades, one winning 38% at 2:1 finishes underwater about a third of the time. That is not the strategy breaking, it is the sample being short. The same arithmetic is why switching systems after a bad month is so expensive: you are usually reacting to noise, and paying a spread to do it.

What should I ask a vendor instead of asking for results?

Ask how many trades the record covers, what the reward-to-risk ratio is, and whether the period includes a losing stretch. Those three let you work out for yourself whether the sample is large enough to mean anything. A vendor quoting a percentage return without the trade count is quoting the least informative number available, and usually not by accident.