QuantForum

Overfitting checklist — 10 signs your strategy is curve-fit

@ea_architectjoined Mar 23, 2025Jul 13, 2026864 views3 replies

After 8 years of building EAs, this is my personal checklist. Score 3 or more and the strategy is probably noise, in my experience:

  1. More than 4 optimized parameters
  2. Parameter "islands" — only one input combination is profitable
  3. Profit factor above 3 on the in-sample test
  4. Perfect equity curve with under 5% drawdown
  5. Works on only one pair and one timeframe
  6. Filters that remove exactly the losing trades
  7. Optimization window shorter than 3 years
  8. No untouched out-of-sample period
  9. Entry logic with more than 5 conditions
  10. You "fixed" the strategy after seeing the first backtest

What would you add or remove?

3 Replies

  1. @quant_newbiejoined Oct 6, 2025#1 · 3 weeks ago
    0

    Number 6 got me. I once added a news filter that removed 11 trades — all losers. I genuinely thought I was a genius. The strategy then lost on the very next news event, live.

  2. @chai_quantjoined Sep 18, 2025#2 · 3 weeks ago
    0

    Add #11: the edge disappears if you shift the entry by one bar. Real edges are robust to small perturbations in timing and parameters. Curve-fits are fragile to everything except the exact test that produced them.

  3. @risk_firstjoined May 23, 2025#3 · 3 weeks ago
    +1

    PF above 3 in-sample should honestly be #1 on the list. In live trading, a sustained PF around 1.8 is elite. Anything above 3 in a backtest is a red flag, not a brag. Print this thread and tape it above your monitor.

Log in or create an account to join the discussion.