Definition
Backtesting simulates how a strategy would have performed using historical data. It is essential for strategy development but fraught with pitfalls: overfitting (finding patterns in noise), survivorship bias (ignoring delisted stocks), look-ahead bias (using future information), and transaction cost underestimation. Robust backtesting uses out-of-sample data, walk-forward analysis, and realistic assumptions.
lightbulb Example
A momentum strategy is backtested on 20 years of data using 15 years for in-sample development and 5 years for out-of-sample validation. In-sample Sharpe is 1.2; out-of-sample Sharpe is 0.7. The decay is expected due to overfitting, but 0.7 still suggests genuine edge.
A strategy that looks great in backtesting may fail live due to overfitting. Always validate on out-of-sample data and use realistic transaction cost assumptions.
verified_user Key Points
- Tests strategy on historical data
- Major pitfalls: overfitting, survivorship bias, look-ahead bias
- Out-of-sample testing validates in-sample results
- Transaction costs and slippage must be realistically modeled