大阪大学　数理・データ科学教育研究センター
Center for Mathematical Modeling and Data Science,The University of Osaka

The Financial Impact of Non-Financial Beliefs in Large Language Models

Dragon Yongjun Tang (The University of Hong Kong)

1. ホーム
2. セミナー・集中講義
3. セミナー
4. The Financial Impact of Non-Financial Beliefs in Large Language Models

大阪大学 数理・データ科学セミナー 金融・保険セミナーシリーズ 第158回

The Financial Impact of Non-Financial Beliefs in Large Language Models

Dragon Yongjun Tang (The University of Hong Kong)

We examine whether large language models (LLMs) encode systematic value-laden beliefs and whether these beliefs shape financial market judgments. ESG issues provide a desirable setting because they combine non-financial preferences with financial materiality, and structured surveys enable direct comparisons between AI beliefs and those of professional and retail investors. Relative to human investors, LLMs assign greater financial relevance to ESG performance, expect larger premia for firms with stronger ESG performance, and are more willing to sacrifice returns for ESG improvements. This orientation stems mainly from post-training alignment and can be attenuated through targeted system prompts. Using analyst reports and broker-level AI rollouts, we further show that analyst optimism increases with firms’ ESG performance after analysts adopt LLMs. Their earnings forecasts for these firms become less accurate than for other firms, consistent with increased bias rather than improved information processing. Overall, non-financial beliefs embedded in LLMs can shape human judgments, revealing a new channel through which AI adoption may affect financial markets.