Does AI substantially improve human decision-making in financial markets?

Determine whether deploying artificial intelligence systems, specifically large language models used for financial statement analysis, can substantially improve human decision-making in real-world financial markets. Assess the practical impact of integrating LLM-generated analyses on decision quality and outcomes for investors and analysts in live market settings.

Background

The paper investigates whether LLMs can perform financial statement analysis using only standardized, anonymized balance sheets and income statements to predict the direction of future earnings. GPT-4 with chain-of-thought prompting outperforms human analysts and performs on par with specialized machine learning models, and GPT-based signals yield profitable trading strategies.

Despite these strong results, the authors caution that translating model performance into improvements in human decision-making requires evidence from practical, real-world usage. They explicitly flag as an open question whether AI can substantially improve human decision-making in financial markets, leaving this issue to future research.