Impact of LLMs on trading decisions and automation of model development

Determine the extent to which large language models will impact future trading decisions and ascertain whether complex tasks—specifically, the development of models for financial analysis—can be automated within agentic AI workflows.

Background

The paper investigates whether LLMs can go beyond sentiment or trend analysis by automating a principled model-building step for financial time series. The authors propose an agentic system that discovers stochastic differential equations and uses the resulting risk metrics to inform trading decisions.

While the paper provides evidence of improved trading outcomes using model-informed agents, the broader question of LLMs’ overall impact on trading and the feasibility of automating complex tasks like model development is framed as an open question in the introduction.