非财务信息的文本分析与反馈：刻画、评价与披露改善

Non-financial information is the textual information which is contained in standard financial report but not in standard accounting statement. It covers contents that are most readable and relevant to decision making. The manual examination of non-financial information, however, has the setbacks of high cost, low precision, and low credibility, making it difficult to extend the research in this field. Thus, this project aims to use the computer-aided textual analysis approach, from the angle of computational linguistics, to quantitatively analyze the disclosure of non-financial information for listed firms in China, and, based on the economic consequences of non-financial information disclosure, further examine the quality of information obtained from computer-aided textual analysis. Next, this project will study the factors that can impact the quality of non-financial information disclosure, analyze the need from online media for non-financial information disclosure, and investigate the possible means to improve the quality of non-financial information. The objective is to provide policy advices to improve the disclosure of non-financial information. Through the study on the eco-system of non-financial information, this project will help create the mechanism to motivate self-disclosure of information, and also improve the disclosure quality. Moreover, through the study on the measurement and estimation of the influencing factors and market consequences of non-financial information disclosure, this project will advance the research of economic theory in China. As a result, it will provide a deeper understanding on the information disclosure in the capital market of China, which will in turn provide the theoretical and empirical support for regulators in making appropriate policy choice.

非财务信息以文本形式呈报了财务报告中会计报表以外的信息，包含了决策中最具可读性和相关性的内容。用手工解读非财务信息存在成本高、主观性大、结论可重复性低的缺点，难以继续拓宽该领域研究的深度和广度。本项目因此将从计算语言学的视角，借助计算机文本分析手段，量化中国上市公司非财务信息披露的内涵，并基于非财务信息披露的经济后果，实证检验计算机文本分析的信息含量。同时，项目拟对影响非财务信息披露质量的诸多因素，以及网络舆情的影响展开研究，分析改善非财务信息质量的着眼点，进而提出上市公司非财务信息披露改善的相关政策建议。本项目对非财务信息系统整个生态空间的研究，有利于激活市场自发的披露激励机制，改善披露质量；而对非财务信息度量、评价以及影响因素和市场反馈的研究，将丰富经济学基础理论，有利于理解中国资本市场的披露现状，为监管层进行政策选择提供理论和实证支持。

利用计算语言学量化文本信息，定性分析语气、管理者特征，风险、竞争以及可读性等数字信息不能报告的内容。本项目发现（1）机器学习迭代优化的文本处理方法优于经典线性预测手段；（2）文本信息的重要市场价值在欺诈预警、违约风险判定、研发决策和财务困境识别领域得到充分的证实；（3）在文本评价指标中，准确率、灵敏度、特异性和精确度的使用频率最高的文本特征度量；此外，本研究还讨论了非财务信息的文本分类在信息检索、信息过滤、情绪分析、推荐系统、知识管理、文档管理等领域都有广泛的应用前景。