Discussion Papers 2025

CIRJE-F-1248	"Investment with New Sentiment Analysis in Japanese Stock Market: Expert Knowledge Can Still Outperform ChatGPT"
Author Name	Lin, Zhenwei, Masafumi Nakano and Akihiko Takahashi
Date	March 2025
Full Paper	PDF file
Remarks

Abstract
This paper presents a novel approach to sentiment analysis in the context of invest- ments in the Japanese stock market. Specically, we begin by creating an original set of keywords derived from news headlines sourced from a Japanese nancial news plat- form. Subsequently, we develop new polarity scores for these keywords, based on market returns, to construct sentiment lexicons. These lexicons are then utilized to guide invest- ment decisions regarding the stocks of companies included in either the TOPIX 500 or the Nikkei 225, which are Japan's representative stock indices. Furthermore, empirical studies validate the effectiveness of our proposed method, which signicantly outperforms a ChatGPT-based sentiment analysis approach. This provides strong evidence for the ad- vantage of integrating market data into textual sentiment evaluation to enhance nancial investment strategies.
Keywords: sentiment analysis, text mining, large language models, natural language process- ing, ChatGPT, Japanese stock market, TOPIX 500, Nikkei 225, investment, alpha creation, risk-adjusted returns

Abstract

This paper presents a novel approach to sentiment analysis in the context of invest- ments in the Japanese stock market. Specically, we begin by creating an original set of keywords derived from news headlines sourced from a Japanese nancial news plat- form. Subsequently, we develop new polarity scores for these keywords, based on market returns, to construct sentiment lexicons. These lexicons are then utilized to guide invest- ment decisions regarding the stocks of companies included in either the TOPIX 500 or the Nikkei 225, which are Japan's representative stock indices. Furthermore, empirical studies validate the effectiveness of our proposed method, which signicantly outperforms a ChatGPT-based sentiment analysis approach. This provides strong evidence for the ad- vantage of integrating market data into textual sentiment evaluation to enhance nancial investment strategies.

Keywords: sentiment analysis, text mining, large language models, natural language process- ing, ChatGPT, Japanese stock market, TOPIX 500, Nikkei 225, investment, alpha creation, risk-adjusted returns