基于变精度粗集神经网的企业失败风险预测研究  

Prediction of Companies' Failure Risk Based on Various Precision Rough Neural Networks

作　　者：尹鹏[1] 王宗军[1] 肖德云[2] 

机构地区：[1]华中科技大学管理学院,武汉430074 [2]武汉理工大学经济学院,武汉430070

出　　处：《管理科学》2010年第4期15-26,共12页Journal of Management Science

基　　金：国家自然科学基金(70872033)~~

摘　　要：描述粗糙集、变精度粗糙集和神经网的相关概念和应用机理,在此基础上构建基于变精度粗糙集的强耦合粗集神经网络,即变精度粗集神经网模型,并以此作为预测中国上市企业失败风险的研究方法。借助信息熵和T检验等手段对财务指标和非财务指标进行筛选,建立企业失败风险预测评价指标体系。在时间跨度上,选择(t-3)年作为研究起点,避免人为夸大预测精度。将被特别处理(ST)的上市企业作为界定企业失败的标准,以160家ST企业和160家配对非ST企业作为测试样本,利用FUSINTER技术离散化相关数据,运用粗糙集、变精度粗糙集、粗糙集神经网和变精度粗集神经网等模型进行实证研究。分析结果表明,在由78家ST和78家非ST企业组成的检验样本中,变精度粗集神经网的预测准确率要高于其他模型,且所需训练时间更短,生成规则更少,训练的整体效果最为理想,由此可以认为该模型比较适用于对中国上市企业进行失败风险的预测研究。Described the relevant notation and mechanism of the Rough Sets Theory （RST）, Various Precision Rough Sets （VPRS） and Neural Networks （NN）, introduced an advanced model-Various Precision Rough Neural Networks （VPRNN） based on VPRS and NN model, and utilized it as a research method for predicting the Chinese listed corporates＇ failure risk. Aapplying the information entropy and T-test to filter various financial and non-financial indicators, consequently, we constructed the enterprise failure risk prediction indicator system. In order to avoid exaggerating prediction result, we chose the data of year （t -3） as the starting point. Meanwhile, considering the Special Treatment （ST） company as the failure enterprise, we selected 160 ST listed companies and 160 non-ST matched companies as test samples and discretized all the data by FUSINTER. Applying the RST, VPRS, RNN and VPRNN models to predict 78 examine samples, we can conclude that the VPRNN model＇s accu- racy is higher than other models, moreover, the training time needed as well as the rules generated is shorter and less than other models. Therefore, it can be interred that the VPRNN model could be more useful and suitable for the prediction of the failure risk for Chinese listed companies.

关 键 词：中国上市企业 风险预测 变精度粗集神经网模型 FUSINTER数据离散化法 

分 类 号：F272.1[经济管理—企业管理]