YannickKae / Statistical-Learning-based-Portfolio-Optimization

This R Shiny App utilizes the Hierarchical Equal Risk Contribution (HERC) approach, a modern portfolio optimization method developed by Raffinot (2018).

Statistical Learning based Portfolio Optimization

Exemplary Hierarchical Asset Structure

This R Shiny application utilizes the Hierarchical Equal Risk Contribution (HERC) approach, a modern portfolio optimization method developed by Raffinot (2018). It combines the unique strengths of the pioneering Hierarchical Risk Parity (HRP) method by López de Prado (2016) and Hierarchical Clustering-Based Asset Allocation (HCAA) method by Raffinot (2017).

Traditional portfolio optimization suffers from significant instability, primarily due to modeling the vector space of return series as a fully connected graph, where each node can potentially substitute for another. This complicated structure magnifies minute estimation errors, leading to unstable solutions. Hierarchical clustering-based tree structures address this issue by eliminating irrelevant links.

As far as I know, there is no other correct implementation of this methodology in R.

User Input

• Use the Ticker table above to enter the securities you wish to invest in. The data is sourced from Yahoo Finance.
• Specify the range of clusters if you wish to narrow this parameter. If not specified, the Silhouette method will determine the optimal number.
• Choose the method to weight the clusters in relation to one another and the assets within each cluster. Risk is defined here as volatility.

Output

• Dendrogram illustrating the hierarchical structure of the securities.
• Pie chart representing the optimized portfolio.
• Table listing each security's portfolio weight and cluster membership.
• Graph depicting the cumulative returns of the securities and the optimized portfolio based on the used data.
• Bar chart comparing the Sharpe ratios of the securities and the optimized portfolio.

The optimal linkage criterion is estimated based on the agglomerative coefficient.

Feature Pipeline

As soon as I find the time, I will make the app more visually appealing and add more features. These include the following:

• Possibility to upload your own data as a csv file.
• More risk metrics, such as Conditional Drawdown at Risk (CDaR)
• Risk-affine weighting
• Shrinkage, denoising and detoning possibilities for the dependency matrix
• Further dependency metrics, such as mutual information.
• Possibility to enter short positions.