BERT人工智能模型能识别管理层语调中的机会主义吗?——基于上市公司年报的文本分析  

Can BERT AI Models Identify Opportunism in Management Tone?A Textual Analysis Based on Annual Reports of Listed Companies

作　　者：洪康隆 Hong Kanglong

机构地区：[1]中证指数有限公司,上海200135

出　　处：《证券市场导报》2024年第10期27-37,68,共12页Securities Market Herald

摘　　要：以往研究发现,上市公司管理层存在通过操纵年报语调积极程度掩盖不利信息,误导投资者作出投资决策的“机会主义”现象。本文以2007—2022年我国A股上市公司年度报告“管理层讨论与分析”(MD&A)章节为样本,探讨BERT人工智能模型对管理层语调中机会主义倾向的识别效果。研究发现,BERT模型度量下的管理层语调积极程度能识别管理层的机会主义倾向,并预测下一年管理层的机会主义行为,且其识别效果比传统的词袋法更好。年报文本语气可操纵程度越高,前述识别效果的差异越显著。进一步研究发现,BERT模型能更准确地预测企业未来业绩表现和股价崩盘风险。本文将BERT模型纳入财经文本情感分析领域,为MD&A语调的度量方法提供了新思路。本文的研究结论进一步完善了信号传递理论,不仅有助于报表使用者对上市公司年度报告这类复杂文本进行情感分析,减少管理层操纵文本带来的错误定价,也为监管部门通过年报文本语调识别管理层的机会主义倾向以及使用人工智能大模型助力数字化、智能化监管提供了证据。Previous studies find that the management of listed companies has the phenomenon of“opportunism”by manipulating the positive tone of the annual report to cover up adverse information and mislead investors into making investment decisions.Based on the sample of the“Management Discussion and Analysis”(MD&A)section from annual reports of China’s A-share listed companies from 2007 to 2022,this paper explores the effectiveness of the BERT artificial intelligence model in identifying the opportunistic tendency in management tone.The study finds that the positive degree of management intonation measured by the BERT model can identify the management opportunistic tendency and predict management opportunistic behavior in the following year,and its identification effectiveness outperforms the traditional bag-of-words method.The difference in identification effectiveness becomes more pronounced as the degree to which the tone of annual report texts can be manipulated increases.Further research shows that the BERT model can more accurately predict companies’future performance and stock price crash risk.This paper introduces the BERT model into the field of financial text sentiment analysis,which provides a new approach to measure the tone of the MD&A section.The conclusion of this paper further improves the signaling theory,which not only helps report users to conduct sentiment analysis on complex texts such as annual reports of listed companies,reducing the mispricing caused by management text manipulation,but also provides evidence for regulatory authorities to identify the management opportunistic tendency through the tone of annual reports and to use large artificial intelligence models to support digital and intelligent regulation.

关 键 词：管理层语调 BERT模型 机会主义 词袋法 

分 类 号：F270[经济管理—企业管理] F832.5[经济管理—国民经济] TP391.1[自动化与计算机技术—计算机应用技术]