Hierarchy Risk Parity with Sliced Wasserstein Distance

Files

• main.py : Main script to run results.
• finance_data_utils.py : Simple tools for financial data.
• hierarchy.py : Hierarchy clustering tools.
• wasserstein.py : Functions for calculating the Wasserstein Distance.
• ./data : Fundemental data files.
• ./results : Saved results including performances and figures.

Overview

This trial explores enhancing HRP portfolio optimization by using sliced-Wasserstein distance for clustering. Results are preliminary and may not outperform the original method, but further research could improve outcomes.

Data

• Time Period: 2014.01.01 - 2024.05.01
• Stocks: Top 20 market cap stocks in the US and Taiwan stock market.
• Rebalance Frequency: 12 months

Methodology

Hierarchy Risk Parity

• Base line model for this asset allocation expieriment.
• Cluster based only on closed price data.

Approximate Sliced Wassestein Distance

• A method that speeds up the calculation of Wasserstein Distance.
• Substitute the distance measure in the hierarchy clustering frame- work with Wasserstein distance measure.

Cluster Results

Cluster with Euclidean Distance

Cluster with Wasserstein Distance

Backtest Results

Taiwan Market 24 Month rolling window

Taiwan Market 60 Month rolling window

Us Market 24 Month rolling window

Us Market 60 Month rolling window