How many trades does a backtest need before the result means anything?

The core idea, without the statistics

Every backtest result is an estimate. An estimate from 30 observations is uncertain; an estimate from 3,000 is less so. The amount of uncertainty depends mostly on the number of observations, not on how impressive the number looks.

This is why a strategy showing an excellent ratio over 40 trades should be treated as a question rather than an answer. Over a small number of trades, a wide range of outcomes is produced by chance alone — including excellent ones.

A quick sanity check you can run today

Take your number of trades. If your strategy were a coin with no edge at all, what range of outcomes would you expect over that many trades? The width of that range shrinks roughly with the square root of the count.

That means going from 100 to 400 trades halves your uncertainty. Going from 100 to 3,000 reduces it by a factor of about five and a half. The improvement is real but it is not linear, and it is slow.

If your reported result sits comfortably inside the range chance would produce, the backtest has not distinguished your strategy from luck. That is a fact about the sample, not an insult to the strategy.

Why this shows up as high ratios on short samples

We keep a record of our own candidates, and the pattern is consistent: the extreme values cluster on the small samples. One of our candidates showed a ratio of 10.0 on 44 observations. That is not a discovery, it is what taking a maximum over many small samples produces.

This is the same mechanism as testing many ideas, applied within a single strategy. A short sample is a small experiment, and small experiments are easy to win by accident.

The trap in the other direction

It is tempting to conclude "more trades is always better". It is not, for two reasons.

First, trades inside a backtest are frequently correlated — a strategy that holds five positions that all depend on the same condition does not have five independent observations. Counting them as five overstates the evidence.

Second, a very large trade count often means a very short holding period, which means the result is dominated by transaction costs. A high count with realistic costs is better evidence than a low count; a high count that only works with unrealistic costs is not evidence at all.

What to do with a report you already have

Look at the trade count first, before the ratio. If it is small, treat every other number as provisional regardless of how good it looks.

Then check the out-of-sample behaviour. A strategy judged on data it was not fitted to is worth more than a strategy with ten times the trades judged on the data that produced it.

Then check the costs. A result that survives realistic spread, commission and swap is a different claim from one that does not.

None of these steps tells you the strategy will work. Together they tell you whether the number in front of you is capable of telling you anything at all.