Assessing portfolio diversification via two-sample graph kernel inference. A case study on the influence of ESG...

Area of Science:

• Quantitative Finance
• Machine Learning
• Network Science

Background:

• Asset and wealth managers require robust frameworks to evaluate portfolio diversification.
• Environmental, Social, and Governance (ESG) investing necessitates selecting assets based on specific screening rules.
• Comparing diversification benefits across different screening rules is a complex challenge.

Purpose of the Study:

• To propose a novel machine learning framework for comparing diversification benefits of portfolios constructed under various screening rules.
• To introduce a method for assessing the influence of ESG screening rules on portfolio diversification.
• To provide asset managers with a tool for evaluating the structural differences in portfolio diversification.

Main Methods:

• Representing screening rules as sequences of graphs, where nodes are assets and edges represent partial correlations.
• Employing a kernel two-sample test, a machine learning hypothesis testing framework, to compare graph sequences.
• Utilizing graph kernels to analyze graph data within the two-sample testing framework.

Main Results:

• The kernel two-sample graph test effectively determines if graph sequences (and thus portfolios) originate from the same distribution.
• Demonstrated the framework's power across various realistic scenarios.
• Applied the methodology to S&P500 data to showcase practical application in asset management.

Conclusions:

• The proposed graph kernel two-sample testing framework provides a statistically sound method for evaluating the impact of ESG screening on portfolio diversification.
• Rejection of the null hypothesis indicates a significant effect of ESG screening on diversification, while failure to reject suggests no significant effect.
• This approach enhances decision-making for asset managers implementing ESG strategies.