Definition
Overfitting occurs when a model is too complex relative to the available data, fitting random noise rather than true signals. An overfit model performs exceptionally in backtesting but fails in live trading. Signs include too many parameters, narrow data period, high in-sample performance with poor out-of-sample results, and lack of economic intuition behind the model.
lightbulb Example
A trading model with 50 optimized parameters achieves a 2.5 Sharpe ratio in backtesting but only 0.2 live. The model was fitting noise in historical data. A simpler model with 5 parameters shows 1.0 backtest Sharpe and 0.8 live—much more robust.
verified_user Key Points
- Captures noise rather than genuine patterns
- More parameters = more overfitting risk
- Test on out-of-sample data to detect
- Simpler models with fewer parameters are more robust