Why a small sample says little
Single trades are noisy. A coin that lands heads six times in ten is not shown to be biased, and a run of ten or twenty-five winning trades is just as easy to get by chance. The smaller the average result is compared with the spread of single trades, the more trades it takes before the average is clearly separate from zero.
The estimate is n = (z / s)^2, where s is the average result per trade divided by its standard deviation. Halving s means needing four times as many trades. The break-even win rate is 1 / (1 + payoff), and the Wilson lower bound shows how low the true win rate could plausibly be given the trades you have.
Costs shrink the average result and so raise the number of trades needed; the cost stress calculator shows how much. To see how much one equity curve depends on its start date, try the random start date explainer.
Frequently asked questions
Where does the number of trades come from?
It uses n = (z / s)^2, where s is the per-trade Sharpe ratio (the average result per trade divided by its standard deviation) and z is how many standard errors from zero you require. With z = 2, the average is about two standard errors above zero. With z = 2.8, the bar is stricter, which is one way to allow for having looked at many ideas.
What is the Wilson lower bound?
A win rate measured on a small number of trades is uncertain. The Wilson score interval gives a range for the true win rate that behaves sensibly for small samples and for win rates near 0% or 100%. The lower bound is the low end of that range. If it sits below the break-even win rate, the sample does not rule out a win rate that loses money.
How is the break-even win rate calculated?
With the average win equal to payoff times the average loss, the break-even win rate is 1 / (1 + payoff). At a payoff of 1 it is 50%; at a payoff of 2 it is about 33%.
Why can the answer be "no trade count"?
If the average trade is zero or negative, collecting more trades will not make a positive result appear, so the formula has nothing to size. If every trade wins, the spread is zero and the formula divides by zero, so it does not apply.
What are the limits of this estimate?
It assumes trades are independent and drawn from the same distribution, and that your inputs are the true values. Real results drift, trades cluster, and inputs measured on a small sample are themselves uncertain. Treat the number as a rough guide to the scale of evidence, not a threshold that proves anything.
Is anything I type stored or sent?
No. The calculation runs in your browser. Nothing you enter is uploaded or stored.