Optimization (Curve-Fitting)

Also called: curve-fitting, over-optimization

Tuning a strategy's parameters to maximise historical performance — risky because it often produces strategies that fit the past but fail forward.

Optimization is necessary but dangerous. The more parameters you tune, the more likely you’re fitting noise rather than signal. Defences: keep parameter count small, use walk-forward analysis, hold out a true out-of-sample period, and prefer parameter robustness over peak performance.