[1O3-GS-10] AI application:

[1O3-GS-10-03] Proposal of a Comprehensive Analysis Framework for Risk Using Text Data of Securities Reports

〇Yusei Karino¹, Daiki Fujiwara¹, Shogo Chomei¹, Yuta Sakai¹, Masayuki Goto¹ (1. Waseda university)

Keywords:Securities report, Risk analysis, BERTopic

A comprehensive understanding of corporate risks is essential for investment decisions. The "Business and Other Risks" section in Securities Reports provides detailed insights into potential risks, but the large volume of text makes cross-company analysis labor-intensive. To improve efficiency, various studies have applied machine learning to risk analysis. However, many lack systematic frameworks, as they do not incorporate pre-existing risk classifications or industry-wide commonalities, limiting the validity of results. This study proposes a method that utilizes 12 risk categories proposed by Noda, based on national risk management standards, to systematically extract risk-related text. By applying BERTopic, we classify industry-common and firm-specific risks, allowing for both a comprehensive overview and a focused understanding of individual corporate risks. An analysis using Securities Reports for Fiscal Year 2023 demonstrates the effectiveness of the proposed approach in enabling systematic and efficient risk identification.