Luck bar calculator
If you try many settings, symbols or EAs and keep the best one, that best one looks good partly by chance. The luck bar is the t-value a strategy with no edge at all is expected to reach when it is the best of the tries you made. A real result has to clear it. This calculator runs in your browser: nothing you type is sent anywhere.
What our own tests taught us, with a trading-cost calculator: What over 11,000 strategy tests taught us.
The luck bar for different numbers of tries
| Tries | Luck bar | Clear with a simple t |
|---|
A worked example
You tried 3 EAs, each on 2 symbols and 2 timeframes, with 8 settings each: 3 x 2 x 2 x 8 = 96 tries. The best one shows 380 trades with an average of +0.12 R and a standard deviation of 1.05 R, so t = 0.12 / (1.05 / √380) = 2.23. The luck bar for 96 tries is about 2.52, so this result is not enough evidence: the best of 96 tries with no edge at all is expected to look about this good. Had it been the only thing you tried, the bar would be 1.96 and 2.23 would be "too close to call" (under 1.25 x 1.96 = 2.45).
How to read it
- Worth a full test: your t is at least 1.25 x the bar. The margin is there because a simple t treats every trade as independent; trades on the same day move together, and in our own demo test the stricter day-clustered standard error was 1.16 to 1.22 times the simple one (measured on that test, not a universal constant: your EA's may differ).
- Too close to call: above the bar but under 1.25 x the bar. A day-clustered t, as the free lab uses, may fall below it.
- Not enough evidence: a result this strong is what the best of your tries is expected to reach with no edge at all.
Limits
- The bar assumes the tries were independent. Settings that are close to each other behave alike, so the true number of independent tries is lower than the count; counting every try still errs on the safe side.
- It does not add trading costs, and it cannot see look-ahead or code faults. The free lab checks those.
- This is a screening aid, not a formal test and not proof of an edge. Clearing the bar is not a promise of future results. Not financial advice.