/ Check 1 · Trade count

How many trades do you need to trust a backtest?

Short answer: it depends on the gap between the win rate and break-even, not on a magic number like 30 or 100. A wide edge shows up in a few dozen trades; a thin one needs hundreds.

Why “30 trades” is the wrong rule

Thirty is a rule of thumb from statistics classes, where it marks the point a normal approximation starts to behave. It says nothing about whether a strategy has an edge. A backtest with 18 wins in 30 trades shows a 60% win rate, but the 95% interval on that count runs from 42.3% to 75.4%. If wins and losses are the same size, break-even is 50%, and that interval still reaches below it. Thirty trades have not shown whether this strategy makes or loses money.

A worked example: same win rate, different answers

Take that same 60% win rate and change only the payoff (average win ÷ average loss):

  • Payoff 1 (wins and losses the same size): break-even is 50%. The interval clears it at about 97 trades. At 100 trades the range is 50.2% to 69.1%.
  • Payoff 1.5: break-even drops to 40%. The interval clears it at about 24 trades.

Now a thinner edge: a 45% win rate at payoff 1.5, five points above its 40% break-even. That takes about 369 trades. At 100 trades its range is 35.6% to 54.8%, which still includes losing money.

Same question, answers from 24 to 369 trades. The number is set by the gap to break-even, and halving the gap costs roughly four times the trades.

Work it out for your own backtest

Enter the win rate the report shows and its payoff:

Break-even win rate50.0%
Trades needed267
95% interval at that count50.0% to 61.8%

About 267 trades, if the win rate holds at 56.0%. Fewer than that and the 95% range still reaches break-even.

The full calculator, with its assumptions, is at /tools/trades-needed. To put an interval around a win rate you already have, use the win-rate interval calculator.

Three things that quietly shrink the count

Bars are not trades. Ten years of data can produce forty trades. The sample is the number of closed trades, not the length of the test window.

Correlated trades count as fewer. Five related stocks bought on the same morning are closer to one decision than five.

Filters spend trades. A filter that lifts the win rate by removing trades also widens the interval. Check the count after the filter, not before it.

What to ask a report for

Wins and total closed trades (not the percentage alone), and the payoff, so you know which break-even the interval has to clear. If the report splits results by year or market, check the count in each piece. If it gives only a percentage and a date range, the trade count is the first thing to ask for.

This is check 1 of 10

The one-page Backtest Verification Checklist, free. Confirm the email and the PDF opens straight away; one worked check each Tuesday after that.

How these numbers were worked out

Every interval here is a Wilson 95% score interval (z = 1.96) on the stated wins and trades. Break-even is 1 ÷ (1 + payoff). “Trades needed” is the smallest count at which the lower end of the interval sits above break-even, assuming the win rate holds where it is and trades are independent; it is the same search the calculator runs. The figures are arithmetic examples, not results of any strategy, and nothing here is investment advice.

Historical simulations are not live results. Nothing on this site is investment advice.