All discussions
EducationOct 28, 18:28 UTC

Best way to backtest a simple trend filter without overfitting?

LKLeila Karim

I have a 200 EMA pullback entry that looks decent on the last six months. I want to test it properly without curve-fitting. What's your process — walk-forward, out-of-sample slices, or something simpler?

1 reply
  • DODaniel OkaforOct 28, 18:28 UTC

    Split into three-year blocks and only look at the last one after the rules are frozen. If you touch a parameter after seeing it, the test is dead.