 Finance-specific large language models: Advancing sentiment analysis and return prediction with LLaMA 2I-Chan Chiu and Mao-Wei Hung Pacific-Basin Finance Journal, 2025, vol. 90, issue C Abstract: In this study, we develop an AI-driven summarization process for lengthy financial texts to improve our self-trained, finance-specific LLaMA-2 model. This approach allows for precise sentiment analysis, leading to more accurate return predictions on disclosures in the Management Discussion and Analysis sections of 10-K filings. Empirical results indicate that trading strategies based on LLaMA-2 sentiments produce significantly higher buy-and-hold returns (BHRs) compared to those derived from FinBERT (Huang et al., 2023) and traditional models. Furthermore, LLaMA-2 sentiment signals show a strong correlation with cumulative abnormal returns (CARs) and surpass traditional methods in predictive accuracy. The summarization process also enhances traditional models, generating significantly higher BHRs with summarized texts than with full texts. Both BHR and CAR results in our approach show robustness during periods of financial turbulence. These findings underscore the value of generative AI in finance and set a new standard for textual analysis. Keywords: Generative AI; Large language model; AI-driven summarization; Textual analysis; Sentiment analysis; Disclosures (search for similar items in EconPapers) Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:eee:pacfin:v:90:y:2025:i:c:s0927538x24003846 DOI: 10.1016/j.pacfin.2024.102632 Access Statistics for this article Pacific-Basin Finance Journal is currently edited by K. Chan and S. Ghon Rhee More articles in Pacific-Basin Finance Journal from Elsevier |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.