Backtesting and overfitting

Level: Advanced 6 minutes · Last reviewed on Oct 13, 2026

A checks a rule on past prices. The danger: if you tweak the settings long enough, you almost always find a curve that looks good – even in pure chance. This is called . A test only becomes honest with data that was not used for tuning (), or with : learn, test, move on. The price series here is pure chance – every “edge” is luck.

Things to try

  • Turn the sliders until the learning phase looks as good as possible. Then show the new data.
  • Press “Find best setting” and compare the learning phase with the new data.
  • Start the walk-forward and roll a new random series.

Quick check

Backtest your own list: backtest with Premium → (with account)

↑ ↓ select · Enter open · Esc close