From Regression to Reasoning: Predicting M&A Announcement Returns With Large Language Models
A new study finds that large language models like OpenAI's o3, GPT-5, and GPT-5.1 can predict short-term stock market reactions to major U.S. mergers and acquisitions better than traditional methods. These AI models, which analyze complex deal and economic data from 2012 to 2022, also help build investment portfolios with higher risk-adjusted returns. The research signals a shift in financial analysis but notes unanswered questions about transparency and performance in other markets or smaller deals.
What it examines
This paper explores if large language models (LLMs) can predict short-term stock market reactions to M&A announcements. Using recent OpenAI models and data from major U.S. deals (2012--2022), the study compares LLM predictions to traditional methods, aiming to improve financial forecasting in M&A.
What it concludes
LLMs outperformed standard models in predicting M&A announcement returns and created better-performing investment portfolios. This suggests generative AI can enhance financial decision-making and portfolio management. Potential uses include supporting investors, analysts, and firms in M&A planning. Future research could address limitations and expand to other financial events.
Evidence objects
Large language models like OpenAI's o3, GPT-5, and GPT-5.1 can accurately predict short-term stock market reactions to major U.S. M&A announcements, outperforming traditional methods such as logistic regression.
key_findings bullet 1 · key_findings · validation V0
The study shows reasoning-based LLMs interpret complex deal, company, and economic data from 2012-2022, enabling investment portfolios with higher risk-adjusted returns and signaling a major shift in financial decision-making.
key_findings bullet 2 · key_findings · validation V0
Despite impressive results, the paper notes limited discussion on transparency and performance in other markets or smaller deals, raising questions about broader applicability and real-world challenges for generative AI in finance.
key_findings bullet 3 · key_findings · validation V0
This paper uniquely applies state-of-the-art LLMs (OpenAIs o3, GPT-5, GPT-5.1) to M& A announcement return prediction, surpassing traditional logistic regression and naive benchmarks in both predictive accuracy and portfolio performance. Its reasoning-based approach to M& A events offers compelling innovation, practical relevance, and significant impact for empirical finance and investment management.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
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.
Source row: 938 · abstract type: unknown