Validation
Strategy Optimization
Strategy optimization compares parameter choices under a consistent test. Its purpose should be to find stable behavior, not merely the highest historical return. Out-of-sample and walk-forward tests are separate validation steps.
Search for stable regions
A broad plateau of similar outcomes is generally more informative than one isolated best parameter. Sensitivity analysis shows how quickly results deteriorate when assumptions change.
Reserve unseen data
Do not use the same observations to select parameters and claim validation. Keep a final period untouched or use a predefined walk-forward schedule.
Penalize unnecessary complexity
Every added rule or parameter creates another chance to fit noise. Compare simpler alternatives and account for the number of trials performed.
Research checklist
- Parameter stability
- Out-of-sample performance
- Number of trials
- Economic rationale
Questions and answers
How do I avoid overfitting?
Limit complexity, predefine the search, inspect parameter stability, reserve unseen data, and require the idea to make economic sense.
Does optimization guarantee future returns?
No. Optimization describes historical sensitivity and can amplify data-mining bias; it does not guarantee future performance.