[3K4-GS-10] AI application: text analysis

[3K4-GS-10-03] Text Mining of Dividend Policy Prospect from Annual Securities Reports

〇Kaito Takano¹, Tomoki Okada¹, Yusuke Shimizu¹, Kei Nakagawa¹ (1. Nomura Asset Management Co., Ltd.)

Keywords:Annual Securities Reports, Dividend Policy, Text Mining, Machine Learning, BERT

Annual securities reports contain various information such as corporate performances, business risks, and ESG activities. In addition to the above information, the annual securities reports also contain dividend policies, which describe not only the basic dividend information for the current year but also future dividend policy and dividend payout ratio. It is well-known that the stock prices move upward significantly after announcements of dividend increases in the future. Although securities analysts predict future dividends based on annual securities reports, they mainly focus on large companies due to the limited resources. In this research, we try to extract information that is useful in predicting future dividends from annual securities reports. We propose a data augmentation method using topic model and BERT model to extract the information with higher accuracy. As a result, we can extract the information of future dividends with high accuracy. For further study, we expect to apply our method to construct new investment criteria and strategies.