Portfolio Optimization Framework

A modular Python framework for comparing covariance matrix estimators and portfolio allocation algorithms using rolling out-of-sample backtesting.

The project is designed to evaluate the impact of covariance estimation on portfolio construction in a high-dimensional setting, with a focus on Global Minimum Variance (GMV) and Hierarchical Risk Parity (HRP) portfolios.

Features

Covariance estimators

• Sample Covariance
• Ledoit–Wolf Linear Shrinkage
• Quadratic-Inverse Shrinkage (QIS)

Portfolio allocators

• Equal Weight
• Global Minimum Variance (GMV)
• Hierarchical Risk Parity (HRP)

Performance evaluation

• Rolling walk-forward backtesting
• Annualized return
• Annualized volatility
• Sharpe ratio
• Maximum drawdown

Technologies

• Python
• NumPy
• Pandas
• SciPy
• scikit-learn
• CVXPY
• yfinance

Project architecture

flowchart TD
    A([Returns])
    A --> B["Covariance Estimator:
        • Sample
        • Linear Shrinkage
        • Nonlinear Shrinkage"]
    B --> C([Covariance Matrix])
    C --> D["Portfolio Allocator:
        • Equal Weight
        • Global Minimum Variance
        • Hierarchical Risk Parity"]
    D --> E([Rolling Backtest])
    E --> F([Performance Metrics])

Loading

Installation

Clone the repository:

git clone https://github.com/<your-username>/portfolio-optimization.git
cd portfolio-optimization

Install the required packages:

pip install -r requirements.txt

Usage

Run the demonstration notebook:

jupyter notebook notebooks/demo.ipynb

Alternatively, open the repository in VS Code and run

notebooks/demo.ipynb

The notebook is self-contained and can be run from top to bottom.

Repository structure

src/
├── allocators/
├── backtesting/
├── covariance/
├── metrics/
└── utils/

notebooks/
tests/
data/

Benchmark

The framework supports systematic benchmarking of different combinations of

• covariance estimators,
• portfolio allocation methods,
• hierarchical clustering linkage methods,

using rolling out-of-sample evaluation.

Example output

The benchmark produces a comparison table containing

Portfolio	Annualized Return	Annualized Volatility	Sharpe Ratio	Maximum Drawdown
Sample | GMV	...	...	...	...
Linear Shrinkage | HRP (average)	...	...	...	...

as well as cumulative wealth plots for each portfolio strategy.