Ryohsuke Tanaka
Recent regulatory reforms in Japan have mandated the inclusion of sustainability-related information in annual reports, offering institutional investors standardized access to its information. However, the absence of evaluation criteria makes it difficult to assess the usefulness of such information for ESG investment decision-making. This study proposes a method that combines generative AI and factor model to identify and evaluate valuable sustainability disclosures. The approach involves extracting distinctive sustainability-related information from companies with relatively strong stock performance using natural language processing techniques. These contents are then scored by generative AI based on specificity, strategic relevance, and measurability. The scores are incorporated into a factor model to examine their relationship with stock returns. Empirical results show that disclosures related to employee well-being have a statistically significant positive association with stock returns.