Decoding market sentiment: the power of ChatGPT in explaining Bitcoin returns from X data

Abstract

Purpose Our research augments the expanding body of literature concerning the capability of prevalent Large Language Models (LLM) tools in supporting financial professionals. We introduce a framework to leverage ChatGPT to assess market sentiment through the analysis of social media data. Design/methodology/approach We use the LLM models to construct market sentiment indicators based on Twitter tweets and use those indicators to explain Bitcoin return. Findings Our analysis uncovers that sentiment indicators crafted with ChatGPT4o/ChatGPT3.5 significantly affect Bitcoin returns, even when accounting for a broad array of control variables and other pre-established sentiment indicators. Originality/value These insights imply that ChatGPT4o/ChatGPT3.5 could empower financial professionals to discover sentiment information from Twitter tweets that were overlooked by previously introduced sentiment indicators concerning Bitcoin.