基金经理语调、基金收益与投资者行为——基于BERT人工智能模型的检验  

Fund Manager Tone,Fund Return,and Investor Behavior——Test Based on the BERT AI Model

作　　者：沈红波[1] 洪康隆 王锴 SHEN Hong-bo;HONG Kang-long;WANG Kai

机构地区：[1]复旦大学经济学院,上海200433 [2]中证指数有限公司

出　　处：《现代金融研究》2025年第4期20-38,共19页Journal of modern finance

基　　金：国家社会科学基金重大项目“负利率时代金融系统性风险的识别和防范研究”(20&ZD102)的资助。

摘　　要：本文以2014-2023年我国基金年度报告“展望”章节文本为样本,运用BERT人工智能模型和传统的词袋法,分别构建“词藻堆砌”与“真情实感”变量,对比基金经理语调对基金未来收益的预测效果及其对个人投资者行为的影响。研究发现:(1)相比传统的词袋法,BERT人工智能模型度量的基金经理语调更能识别基金经理的真情实感,对基金未来收益、顺境中的基金业绩持续性及逆境中的基金崩盘风险的预测效果更好,且上述识别优势在复杂度较高、可读性较差的“展望”文本中更显著;(2)个人投资者的“有限理性”特征和公募基金存在的委托代理问题导致个人投资者更容易被基金经理“词藻堆砌”的文本所吸引,进而导致投资收益率降低。本文将BERT人工智能模型引入财经文本情感分析,为文本语调的度量方法提供了新思路。Taking the texts of the“Outlook”chapter of the annual report of Chinese funds from 2014-2023 as a sample,this study uses the BERT AI model and the traditional Bag-of-Words(BoW)method to construct“rhetorical embellishment”and“genuine sentiment”variables respectively.It compares the predictive effects of fund manager tone on future fund return and its influence on individual investor behavior.The findings reveal that:(1)Compared to the traditional BoW method,the fund manager tone measured by the BERT AI model is more effective in capturing genuine sentiment.It demonstrates superior predictive power for future fund return,fund performance persistence in favorable conditions,and crash risk in adverse conditions.This identification advantage is more pronounced in complex and less readable“Outlook”texts.(2)The“bounded rationality”characteristics of individual investors and the principal-agent problem in public funds lead individual investors to be more easily attracted by“rhetorically embellished”texts,resulting in lower investment returns.The study introduces the BERT AI model into sentiment analysis of financial texts,providing a novel approach for textual tone measurement.

关 键 词：基金经理语调 基金收益 投资者行为 BERT模型 

分 类 号：F832.48[经济管理—金融学]