公告溢价效应与资产定价:文本机器学习视角  

Announcement Premium Effect and Asset Pricing:A Textual Machine Learning Perspective

作　　者：唐国豪 朱琳 陈世程 TANG Guohao;ZHU Lin;CHEN Shicheng(Hunan University,Changsha,China)

机构地区：[1]湖南大学金融与统计学院,410006

出　　处：《经济学动态》2024年第2期72-87,共16页Economic Perspectives

基　　金：国家自然科学基金青年项目(72003062)。

摘　　要：本文在中国股票市场中,针对上市公司公告文本数据,采用文本分析机器学习方法进行信息提取,旨在揭示公告信息与资产预期回报之间的关系以及对资本市场的影响渠道。本文首先依据监督式训练方法构造了基于公告的情感词典,并以此为基础采用机器学习方法对公告效应进行实证分析,最后从多个渠道探究了公告溢价效应的市场反馈机制。本文研究发现,基于机器学习的公告文本情感能显著预测股票收益率,且明确存在正向显著的公告溢价效应。异质性分析发现,公告效应在小规模、成长型公司中溢价显著;与国企相比,民营企业的公告效应更显著。机制分析发现,公告溢价产生的主要原因可能是由于散户投资者的过度关注。对于金融机构关注较多和信息披露质量较高的公司,公告溢价效应较弱。In recent years,the application of machine learning techniques within the financial market has become increasingly prevalent,underscoring the significant potential of these methods in accurately forecasting and explaining the market returns.Concurrently,big data is revolutionizing the finance industry and has the potential to significantly shape future research in finance.Goldstein et al.(2021)noted that new research built on these big data to push the frontier on fundamental issues across areas in finance.Textual data,with its large size,high dimensionality,and complex structure,epitomizes big data features.Corporate announcements are particularly noteworthy within this domain due to their high‐quality information,which not only reveals a company’s operational and development status but also communicates critical information that can significantly affect stock price.Consequently,it is valuable to mine effective pricing information from these corporate announcements to examine its predictive impact on returns.This paper analyzes textual data from corporate announcements spanning from 2000 to 2020.Through textual analysis,this paper constructs the sentiment dictionaries based on these announcements and employs machine learning techniques to assess the sentiment orientation(positive or negative)of each announcement,so as to investigate the influence of announcement content on the anticipated asset returns.Furthermore,this paper undertakes robustness tests and heterogeneity analyses of the announcement premium effect,offering an in‐depth examination of the market’s reaction mechanisms to this effect from multiple dimensions.The findings of this paper reveal several key insights.Firstly,there is a significant variance in the returns of stocks based on the category of announcements,with positive news yielding higher returns and negative news leading to lower returns.Secondly,the sentiment dictionary developed in this paper effectively differentiates the direction of stock returns,demonstrating a significantly

关 键 词：公告溢价效应 文本分析 机器学习 市场反馈 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.1[自动化与计算机技术—控制科学与工程] F832.51[经济管理—金融学] F275