Explainer
How many trades does a backtest need?
There is no set number: a backtest needs enough trades that one lucky or unlucky stretch cannot decide the answer, so read the count beside every figure.
Updated
The short answer
There is no magic number. How many trades you need depends on what you want to say with the result.
A result from a handful of trades describes those trades. It says very little about the next ones. A result from thousands of trades says much more, because no single trade can move it.
A fair coin shows why. It can land on heads several times in a row. Over thousands of flips it settles close to half. Backtest figures behave the same way: the fewer trades behind them, the more they can move on luck alone.
Read the number of trades next to every figure. A figure without its count tells you nothing about how much weight it can carry.
Why 50 trades and 2,000 trades let you say different things
With 50 trades you can say what happened in those 50 trades. Take out a few and the picture can change. You cannot split them by year and still have enough in each year to look at.
With 2,000 trades, no single trade matters much. You can split the history into years, or into morning and afternoon entries, and still have a real count in each part. You can ask whether the result looks similar across years, and the answer means something.
Our own site puts it this way: fourteen trades over eight months is a number, not a pattern.
- Sample size
- The number of trades behind a figure.
- Interval
- A range around a figure showing how far it could move with a different set of trades. More trades make the range narrower.
Count the spread, not only the total
Trades on the same day, or in the same stretch of market, tend to look alike. Two thousand trades from one calm year are not the same as two thousand trades spread across many years.
Our NQ and ES data runs from June 2, 2016 through August 27, 2026. A rule that can trade on any day has years of days to draw on. A rule that needs a rare setup may produce a small count even over that whole span.
What lowers your trade count
The count you get is almost always smaller than the number of days in the data. These are the usual reasons:
- Filters: every condition you add removes days.
- A narrow entry window. A setup whose window runs out with no entry taken is skipped.
- A bar that opens beyond the stop or the target on the entry bar. The trade is skipped, and the results list it as skipped.
- Held-back months. About 30% of months are kept out of the results, and any trade that is open in one of them is dropped.
What to do with a thin sample
Say plainly that it is thin. Do not round a small sample up into a conclusion.
Then you have options. Loosen a filter so the rule trades more often. Test a longer history. Test the same rule on a second market, written out the same way. Or keep the rule as it is and wait for more trades to arrive.
None of this makes a bigger sample a promise. More trades make the figure steadier for the history you tested. They do not say what happens next.
Questions and answers
Is 100 trades enough for a backtest?
There is no cutoff we can give you. It depends on what you want to say. One hundred trades are enough to describe those trades. They are a thin base for splitting by year or by condition.
Does more history always mean more trades?
Only if the rule trades often. A rule that fires a few times a year adds very little with each extra year. Count the trades, not the years.
Do skipped trades count in the sample?
No. A skipped setup, for example one whose entry bar opened beyond the stop or the target, is listed as skipped. It is not counted as a win or a loss.
Where do I find the trade count?
Next to the figures. Every finding on this site states its sample size beside the results.