Corporate carbon information disclosure and asset mispricing

Owing to the implementation of new accounting standards in China since 2007, we use Python software to capture the relevant disclosures in the social responsibility, sustainability, and environmental reports of China's A-share listed companies from 2007 to 2021 from the Shanghai and Shenzhen Stock Exchanges. We use text analysis technology and natural language processing (NLP) technology to assign scores and then use the weighted sum of indicator weights obtained by the analytic network process (ANP) to finally obtain the data of carbon information disclosure samples. Asset mispricing level data rely on the stock trading volume from the Wind database. Financial data are obtained from the CSMAR and CNRDS databases. Initial data are further excluded as follows: (1) samples from the financial industry, (2) special treatment samples, and (3) samples with missing data. In addition, to reduce the potential interference of outliers, all continuous variables are winsorized at the 1% level. Finally, this study obtained 6,048 firm-year observations.

自2007年以来，随着我国新会计准则的实施，本研究利用Python软件从上海证券交易所和深圳证券交易所捕捉了2007年至2021年间中国A股上市公司在社会责任、可持续性和环境报告中的相关披露信息。本研究采用文本分析技术和自然语言处理（NLP）技术对相关文本进行评分，进而通过分析网络过程（ANP）获得的指标权重加权求和，最终获取碳信息披露样本数据。资产错价水平数据依赖于万德数据库的股票交易量。财务数据来源于CSMAR和CNRDS数据库。初始数据经过如下筛选：（1）金融行业样本，（2）特殊处理样本，（3）数据缺失样本。此外，为了降低异常值可能带来的潜在干扰，所有连续变量均在1%的水平上进行Winsorize处理。最终，本研究收集到6,048个公司-年度观测数据。