How many trades before a win rate means anything?
A 70% win rate over 12 trades and over 400 trades are different claims. Here is the arithmetic that separates them.
Win rate is a proportion, and proportions carry error bars that shrink slowly.
The rough standard error of a win rate p over n trades is:
se = sqrt(p * (1 - p) / n)
Put real numbers in it.
| Trades | Observed win rate | Roughly plausible range |
|---|---|---|
| 12 | 70% | 45% – 95% |
| 50 | 70% | 57% – 83% |
| 200 | 70% | 64% – 76% |
| 500 | 70% | 66% – 74% |
At twelve trades the honest reading of "70%" is "somewhere between a coin flip and a miracle". That is not pedantry — it decides whether the edge survives costs.
Why this matters more on high timeframes
A weekly strategy over two years produces perhaps 20–40 trades. That is not a flaw in the test, it is the amount of evidence two years of weekly candles can physically contain. Either accept the wide range or test more history.
What to do instead of hoping
- Print the trade count next to every rate. Always.
- Treat fewer than ~30 trades as directional, not measured.
- Prefer expectancy per trade over win rate — a 40% win rate at 2.5R beats a 70% win rate at 0.4R.
- Check the result survives on candles the strategy was never tuned on.
A small sample is not a reason to discard a strategy. It is a reason to stop describing it with two decimal places.