Portfolio optimization and backtesting

A Python class for portfolio construction, risk-aware optimization and rolling backtests. Supports empirical mean shrinkage, Ledoit–Wolf covariance, EWMA, CVaR, risk-parity, turnover-aware Sharpe optimization.

See demo Jupyter notebook for use examples.

Features

• Fetch asset prices (Yahoo Finance) and macro series (FRED)
• Pairwise geometric means and covariance estimation with handling of missing data
• Empirical shrinkage of expected returns (James–Stein)
• Ledoit–Wolf covariance shrinkage for stable, well-conditioned covariance
• EWMA covariance for reactive risk modeling
• Portfolio optimizers:
  • Max Sharpe ratio
  • Min-variance
  • Risk-parity (equal risk contribution)
  • CVaR minimization
  • Turnover-aware Sharpe optimization

Backtesting

• Rolling backtest with configurable window and rebalance frequency
• Supports all optimization methods including turnover control
• Returns portfolio-level metrics: annualized return, volatility, Sharpe ratio and maximum drawdown

Statistics & Metrics

• Computes portfolio performance: expected return, volatility and Sharpe ratio
• Generates backtest summary statistics and per-asset weight tables
• Efficient frontier simulation and plotting

Covariance & Mean Estimation

• Nearest PSD cov matrix calculation and diagnostics
• Pairwise covariance with positive-definite adjustment
• Ledoit–Wolf shrinkage (scikit-learn)
• EWMA covariance with configurable halflife
• Empirical Bayes / James-Stein style mean shrinkage towards cross-sectional average

Optimization Methods

• optimize_sharpe – maximize Sharpe ratio
• minimize_variance – minimize portfolio variance
• optimize_risk_parity – equalize risk contribution
• optimize_cvar – minimize Conditional Value at Risk (CVaR)
• optimize_sharpe_with_turnover – penalize high turnover for transaction cost control

Outputs

• Asset allocation weights and expected returns
• Portfolio-level metrics: return, volatility, Sharpe ratio, max drawdown
• Dataframes for easy reporting and analysis

Dependencies

• numpy, pandas, scipy, matplotlib, yfinance, pandas_datareader, scikit-learn