数字技术与经济金融前沿论坛（第50期）

摘要：

The privacy preservations of these algorithms are justified theoretically in the sense of carefully formulated definitions. Furthermore, we demonstrate the applications of these algorithms to machine learning.We introduce a large language model-based sentiment analysis framework (LLM-MESA) that enables fine-grained, entity-specific measurement of sentiment in financial texts. Applying this method to mutual fund manager reports, we extract expectations regarding equities, the macroeconomy, and bonds, providing the first evidence on how institutional stock market sentiment predicts crash risk. We find that managers’ optimistic equity outlooks significantly increase crash risk, indicating that institutional forecasts embed non-rational sentiment. This effect is concentrated among lower-skill funds, when macroeconomic expectations are pessimistic, and when bond market views are optimistic. Mediation analysis shows that optimism amplifies institutional herding and attracts sentiment-driven retail investors, thereby exacerbating crash risk. Our study demonstrates the value of LLM-MESA for financial text analysis, advances understanding of the mechanism linking institutional sentiment to crash risk and contributes to the debate on whether institutional investors stabilize or destabilize markets.

主讲人介绍：

刘嘉，英国管理学院院士，朴茨茅斯大学会计与金融学教授。她在英国伯明翰大学获得博士学位，曾经就职于利兹大学和索菲尔德大学。她的研究包括公司金融、行为金融、金融市场、风险管理、金融科技、 大数据、 会计报告和披露,和可持续性。她在主要金融、会计、管理学国际期刊发表80多篇学术论文, 其中包括近20篇ABS-4和FT50期刊; 并出版关于公司治理、可持续性、和决策方面的专著。曾获欧洲管理杂志 (EU)、英国管理学院(英国)、西南金融协会(美国)、以及Eurasian商业与经济协会(EBES)的“最佳论文奖”。她的研究被多个媒体采访和发表,包括著名的The Conversation, 华盛顿邮报, 以及纳斯达克证券交易所。