Enhancing Portfolio Construction with Correlation Estimates from Large Language Models

This study explores the potential of large language models (LLMs) to enhance portfolio optimization by incorporating their insights into asset correlation structures. The authors propose a novel framework that integrates LLM-derived relationship information into traditional portfolio models using a modified Gerber statistic. The key contribution is a methodology that captures LLMs’ semantic understanding of asset interdependencies and embeds it within a quantitative investment process. Empirical results demonstrate that leveraging GPT’s insights leads to more efficient portfolios, highlighting the practical value LLMs can offer in augmenting modern investment strategies.