Corporate Event Predictions Using Large Language Models

Abstract

This paper offers a thorough assessment of large language models (LLMs) in the context of corporate event prediction. To achieve this, we formally establish the task of corporate event prediction, construct a novel dataset containing summaries of earning call transcripts, and conduct comprehensive experiments involving both raw and fine-tuned variants of the primary LLMs. Our experimental findings underscore the viability of automating this intricate task using LLMs and highlight the unnecessariness of additional finetuning.