基于聚类-粗糙集-神经网络的企业财务危机预警

Early Warning of Financial Distress Using Clustering-Rough Sets-Neural Networks

财务危机预警是企业各方利益相关者关心的一个重要问题。由于样本的可得性问题,现有的财务预警相关研究一般只能把企业的财务状况分为ST和非ST两类,运用层次聚类将其分为健康、良好、中等、轻警、重警5种类型,从而对企业财务状况的描述更符合实际。在考虑预警指标所含信息的完备性和准确性的基础上,使用了粗糙集方法构建指标体系,弥补了传统指标体系构建时信息量冗余或不足的缺陷。最后,将5种分类作神经网络的输出层,粗糙集约简后指标体系作输入层,构建神经网络模型进行财务预警。通过粗糙集约简形成的输入层节点数以及聚类形成的输出层节点数,进而提升了神经网络的拟合效率,使其运用更为合理,并最终完成对财务状况转折点以及财务状况恶化阶段的预测。

Financial crisis warning of listed companies is always an important concern of stakeholders. Due to the data availability, reseachers typically divide companies into two classes as ST and non-ST. Besides, past research pay less attention to the indicator selection, subjective judgement may be the main way for this issue. This paper aims to ovecome the two limitations about finacial situation level and indicator selection. Rough set theory is used to set up a complete and simple indicator system, while hierarchical clustering analysis is used to classify the financial status into five levels, namely health, relative health, medium. Medium, slight warning and serious warning. This changes traditional classification scheme with only ST and non-ST classes. A neural network model is built using the reduced indicator system as the input and the five financial status levels as the output. The model is more accurate in predicting the financial distress status because of more reasonable neural network structure design.