Research on Sentiment Analysis of Japanese Corporate Disclosures Using Large Language Models and Stock Return Predictability by Professor Katsuhiko Okada, Institute of Business and Accounting, Kwansei Gakuin University, is Published in PeerJ Computer Science

On November 18, 2025, the results of research by Professor Katsuhiko Okada, Institute of Business and Accounting, Kwansei Gakuin University, on sentiment analysis of Japanese corporate disclosures using large language models and stock return predictability were published in PeerJ Computer Science.

The research of Prof. Okada demonstrated that advanced large language models (LLMs) such as ChatGPT, Claude, and Gemini can extract sentiment from Japanese 10-K reports that predicts future stock returns—a capability that traditional dictionary-based methods fail to achieve. By analyzing over 70 million words from 11,135 firm-year observations of Tokyo Stock Exchange-listed companies between 2014 and 2023, the study found that stocks identified as having positive sentiment by LLMs tend to underperform subsequently, suggesting that investors initially overreact to optimistic corporate disclosures and the market later corrects this overpricing. This finding challenges the Efficient Market Hypothesis, which posits that all publicly available information is already reflected in stock prices.

The research is expected to advance AI-driven financial analysis and investment decision-making. Future research directions include testing the generalizability of these findings in other international markets and investigating how the predictive power of such signals may evolve as AI adoption in finance becomes more widespread.

Journal name: PeerJ Computer Science
Article title: From words to returns: sentiment analysis of Japanese 10-K reports using advanced large language models
Author(s): Katsuhiko Okada, Moe Nakasuji, Yasutomo Tsukioka, Takahiro Yamasaki
DOI: 10.7717/peerj-cs.3349