Global Merger-Arbitrage Forecasting with Language Models
Hinal Jajal, Michał Mucha, Charles Sweat, Chris Pulman, Charlie Flanagan, Peter Anderson
Abstract
Prior work on judgmental forecasting with large language models (LLMs) has focused on broad, mixed‑topic question banks and shallow context (e.g., short news snippets). We study a specialized, high‑stakes financial setting: forecasting M\&A outcomes for merger arbitrage. Using rich textual evidence, with context engineering informed by veteran merger-arb specialists, we construct an LLM‑based forecasting system and finetune the model using outcome-conditioned gold reasoning traces. The system outputs probabilistic forecasts over closing at announced terms, higher bid, and deal termination. On an out-of-sample set of more than 400 large deals spanning 42 countries, our finetuned system outperforms a variety of frontier models and market-based baselines, using a Brier score weighted by the P\&L impact of each deal.
BibTeX
@inproceedings{
jajal2026global,
title={Global Merger-Arbitrage Forecasting with Language Models},
author={Hinal Jajal and Micha{\l} Mucha and Charles Sweat and Chris Pulman and Charlie Flanagan and Peter Anderson},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=wDaX9bFufS}
}