RiskLab researchers from the Universities of Toronto and and Montreal published a paper introducing a novel machine learning (ML) framework for causal discovery based on recent advances in large language models (LLMs) and discuss the applications of these causal discovery techniques to investment management.
Unlike typical data-driven methods for data discovery, the framework using the implicit “world knowledge” in state-of-the-art LLMs to automate the expert judgement approach to causal discovery. A key application that is explored in detail is end-to-end causal factor analysis, where the authors demonstrate the utility of the method in specifying and analyzing detailed causal models for financial markets.
Factor investing, which involves identifying and leveraging specific factors or drivers behind asset returns, serves as a robust testbed to showcase the efficacy of the framework. The finance industry, characterized by its complex and dynamic data, demands robust methods for causal analysis to enhance investment strategies and risk management. Unlike conventional correlation-based analyses, this paper looks at creating alternate systematic approaches for selecting factors for constructing factor models.
The paper also conducts a comparative analysis, juxtaposing the new approach with conventional methods, to underscore the enhanced capability of the framework in revealing intricate causal dynamics in financial data.