This research investigates whether large language models like ChatGPT and DeepSeek can extract predictive information from financial news sources, specifically the Wall Street Journal, to forecast stock market movements and macroeconomic trends. The study represents a significant contribution to the emerging field of AI-driven financial analysis by systematically evaluating the capabilities of different LLMs in processing economic news and generating actionable market insights.

The methodology involves analyzing Wall Street Journal content using multiple large language models, with ChatGPT demonstrating superior performance compared to DeepSeek and other models. The researchers find that ChatGPT's predictive power aligns with established financial theories, particularly regarding investor underreaction to positive news during challenging economic conditions. The study reveals that while negative news correlates with market returns, it lacks predictive value, whereas positive news during periods of economic downturn and high information uncertainty drives the observed predictability. The research concludes that ChatGPT currently stands alone among tested models in its ability to capture economic news information that connects to market risk premium, highlighting its potential as a tool for financial analysis while noting limitations of other LLMs in this domain.