What overfitting looks like
Overfitting happens when a strategy matches noise instead of signal. It looks great on a specific slice of history, but collapses elsewhere. Common warning signs:- extreme results on short ranges
- performance collapses out of sample
- tiny parameter changes break the strategy
- the logic becomes complex without a clear reason
Why it happens
Overfitting is usually caused by:- too many degrees of freedom (filters, conditions, parameters)
- too little data for the number of choices
- repeated tweaking until results look good
- optimizing to a single market regime
Guardrails that actually help
Use these to keep strategies realistic:- Reduce complexity before optimizing. Fewer moving parts = fewer false positives.
- Test across ranges and regimes, not just one period.
- Prefer logic changes over parameter tweaks when behavior is wrong.
- Require stability, not peak performance. Consistent results beat a single perfect run.
- Compare to a baseline so you know what really improved.
A strategy that is slightly worse but stable is usually better than a perfect, fragile one.