Who this is for — Anyone building or refining strategies who wants to avoid the classic mistake: confusing a pretty curve with a solid method.
Overfitting happens when rules and parameters are fitted too closely to the details of the historical sample. Results look excellent in test but lose reliability as soon as conditions and market noise change.
In plain terms — You "memorised" the past, not the principle that generates it. At the first different question, the system wobbles.
Bronze prerequisite — Before this lesson: trading-journal, trade-result, trading-mistake. See bronze-path.
Typical signs of overfitting
Some clues recur often:
- too many filters and parameters for a unclear edge;
- performance very sensitive to small changes;
- large gap between in-sample and out-of-sample;
- logic hard to explain without reference to the historical chart.
Complexity is not bad in itself, but it must be justified by real robustness.
Example — Strategy with 11 conditions produces perfect equity over 5 years. Removing two secondary filters, results collapse: likely dependence on historical coincidence, not structural edge.
How to reduce it in practice
- Start from simple, verifiable rules.
- Limit optimisations and number of free parameters.
- Validate on separate periods with out-of-sample.
- Stress the system with robustness tests.
Card
- What it is: excessive adaptation of the strategy to data used to develop it.
- When to use it: every time you optimise rules, filters, or parameters.
- Typical mistake: chasing maximum historical profit without asking whether it is repeatable.
Silver path — Module: Validation. Part of silver-path.