How many trades
is enough?
A win rate without a sample size is a slogan. Enter the wins and the total trades, and this page shows the range the true win rate could still be in — plus the trade count that would make the result evidence.
Test a result
Count trades, not days. A series of daily returns is a different test — use the Lie Detector for prediction files.
| At this win rate | Trades needed |
|---|
"Min. trades at this rate" is the smallest sample where the lower bound of the Wilson interval rises above 50%. It is conservative: it demands the whole interval clear a coin flip, not just the point estimate.
Enter a result and press Calculate. Nothing is sent anywhere — the arithmetic runs in this tab.
What the number means
Suppose a vendor reports 70% from 20 trades. The point estimate is 0.70. The Wilson 95% interval is roughly 48% to 86%. A coin flip sits inside that band, so the sample cannot tell skill from luck. The headline is real; the evidence is not.
The same 70% from 600 trades narrows to roughly 66% to 74%. Now the whole interval sits above 50%, and the claim survives. Nothing about the strategy changed. Only the sample did.
Why the Wilson interval, not the textbook one
The normal approximation to a proportion breaks down when the sample is small or the rate is near 0% or 100%. It can even produce bounds below zero or above one. The Wilson score interval stays inside 0–100%, behaves sensibly at small n, and is the same formula the BuildIT certifier uses on every accuracy figure.
The trap after the interval
A wide interval is a reason to collect more data, not a reason to reject the strategy. But more data only fixes sampling error. It does not fix overfitting, look-ahead bias, survivorship bias, or a model tuned on the same rows it was scored on. Sample size is necessary and not sufficient.
That is why the free Lie Detector runs baselines and walk-forward folds as well as the interval. A number can be precise and still be wrong.
Questions
How many trades do I need for a valid backtest?
It depends on the win rate. At 95% confidence, a 70% rate needs about 25 trades before the lower bound clears 50%, a 60% rate about 97, and a 55% rate about 385. Treat those as floors: at the floor, the bound only just clears 50%. Below 100 trades, treat the result as a hypothesis, not evidence.
Does this work for accuracy instead of win rate?
Yes. Accuracy is a proportion, so the arithmetic is identical. Enter correct predictions as wins and total predictions as trades.
What if the win rate is below 50%?
The calculator will report that no sample size clears 50%, because the interval is centred below it. A strategy can still be profitable below a 50% win rate if winners are larger than losers — but that is a question about expectancy, not win rate. Use the prop-firm simulator for that.
Is this investment advice?
No. It is arithmetic on a claimed sample. It does not predict returns, and a narrow interval around a backtested number says nothing about the next market regime.