Stock Return Prediction with SSESTM Model using Quarterly Japanese Company Handbook

Abstract

In this study, we perform an empirical analysis of text-mining methodology that extracts sentiment information to predict stock returns. We use the Quarterly Japanese Company Handbook, which is a widely-acclaimed quarterly publication on the Japanese stock exchange, but there are few studies using it. As for the sentiment analysis model, we focus on the Supervised Sentiment Extraction via Screening and Topic Modeling (SSESTM). It has been proposed as a sentiment analysis specialized for stock return forecasting and produced a substantial profit in the U.S. stock market. The SSESTM using the stock return as a teacher label, but we propose using the specific return. The stock return can be decomposed into various common factors such as market and size, and firm-specific return. The Quarterly Japanese Company Handbook provides the comments of companies' earnings forecasts, and it is considered more useful for forecasting specific returns than stock returns including common factors. We examine for prediction the specific return in the Japanese market using Quarterly Japanese Company Handbook. As a result, we confirm that the SSESTM model using four years of articles in the training data gave relatively good results for the high quantile stock groups, but not for the low quantile stocks.