Publication: Intraday stock prediction using sentiment analysis: evidence from dividend announcements
Authors
Álvarez-Diez, Susana ; Baixauli Soler, J. Samuel ; Kondratenko, Anna ; Lozano Reina, Gabriel
item.page.secondaryauthor
item.page.director
Publisher
Taylor and Francis Group, Routledge
publication.page.editor
publication.page.department
DOI
https://doi.org/10.1080/15427560.2025.2538879
item.page.type
info:eu-repo/semantics/article
Description
© 2025. This document is the Accepted version of a Published Work that appeared in final form in Journal of Behavioral Finance. To access the final edited and published work see https://doi.org/10.1080/15427560.2025.2538879
Abstract
This study explores whether sentiment extracted from financial news using large language models (LLMs) can predict abnormal intraday stock returns following dividend announcements. Drawing on 4,682 news items linked to 1,258 announcements from 394 S&P 500 companies (January 2023–January 2024), we use ChatGPT to extract sentiment polarity scores and we apply different models to forecast cumulative abnormal returns (CARs) in 30-minute intervals. Our findings reveal that sentiment – especially when captured immediately after news releases – has significant predictive power over intraday price movements. Strategies based on ChatGPT-derived sentiment consistently outperform benchmark models, particularly within the first two hours of trading. These results remain robust across alternative specifications and placebo tests, highlighting the value of LLMs for real-time market prediction. This research advances the literature on sentiment analysis and behavioral finance by linking emotion-driven news interpretation to high-frequency trading performance.
publication.page.subject
Citation
Journal of Behavioral Finance, 2025
item.page.embargo
2027-01-31
Collections
Ir a Estadísticas
Sin licencia Creative Commons.