Overview

A robust backtesting framework is essential for validating strategies before live trading.

Key Principles

1. Avoid look-ahead bias - Use only information available at decision time
2. Realistic costs - Include all transaction costs and slippage
3. Survivorship bias - Use point-in-time universes
4. Out-of-sample testing - Reserve data for final validation

Simple Backtest Example

See Code Examples for a complete implementation of SimpleBacktest class.

Key features:

• Daily rebalancing
• Transaction cost modeling
• Position tracking
• Performance metrics

Performance Metrics

Returns-Based

• Total return
• Annualized return
• Volatility
• Sharpe ratio
• Maximum drawdown
• Calmar ratio

Risk-Adjusted

• Information ratio (vs benchmark)
• Sortino ratio
• Omega ratio

Factor Analysis

• Alpha and beta vs benchmarks
• Factor exposures over time

Common Pitfalls

Data issues:

• Using adjusted prices incorrectly
• Not handling splits/rolls properly
• Missing or incorrect data

Implementation issues:

• Trading at close prices (look-ahead)
• Ignoring partial fills
• Underestimating costs

Statistical issues:

• Overfitting to in-sample data
• Multiple testing (p-hacking)
• Ignoring regime changes

Walk-Forward Analysis

Best practice: rolling window optimization and testing.

1. Train on window 1
2. Test on window 2
3. Slide forward and repeat
4. Aggregate out-of-sample results

Typical windows:

• Training: 2-3 years
• Testing: 6-12 months
• Step size: 3-6 months

Cross-Validation

For parameter selection, use k-fold cross-validation:

• Split data into k folds
• Train on k-1, test on 1
• Repeat for all folds
• Select parameters with best average performance