From Regression to Reasoning: Predicting M&A Announcement Returns With Large Language Models

ABSTRACT This study investigates whether large language models (LLMs) can predict short‐term market reactions to M&A announcements. We prompt OpenAI's latest reasoning models (o3, GPT‐5, and GPT‐5.1) to forecast whether the combined market value of acquirer and target will increase or decrease, drawing on deal‐, firm‐, and macroeconomic data for large domestic U.S. transactions (2012–2022). Our analysis shows that LLMs outperform logistic regression and naive buy‐all benchmarks in predictive accuracy. Their forecasts further translate into portfolios with superior risk‐adjusted performance. These findings highlight the transformative potential of generative AI for empirical finance and M&A decision‐making.