Overfitting a trading strategy
Overfitting occurs when a strategy is tuned to historical noise rather than reproducible structure. It often looks great in-sample and disappears as soon as you test on unseen data.
The pattern to watch
The clearest sign of overfitting is a parameter set that is excellent in-sample but unstable under small walk-forward changes. A strategy that only works when you choose a precise combination of inputs can be a statistical artifact rather than an edge.
Why walk-forward matters
A walk-forward split keeps the model honest by measuring performance on data it never saw during tuning. If the same configuration only works in one narrow window, it is usually not a durable edge.
What to do instead
Use a simple, clearly defined rule set. Keep the parameter space broad enough to avoid hand-picking a magical setup. Then validate the result on a holdout period and test the same logic across multiple instruments and timeframes.