基于BP神经网络的上市公司财务预警模型

为了进行企业财务危机预警方法精度的比较研究,采用BP人工神经网络工具,以120家上市公司的截面财务指标作为建模样本,并使用同期60家公司作为检验样本建立了财务危机预警模型.经过对样本的反复训练和学习,分别取得了建模样本90.8%和检验样本90%的判正率.与我们采用主成分分析法建立的模型对同一建模样本和检验样本的预测精度分别是90%和81.7%相比有很大的提高.研究结果表明:BP神经网络是一种非线性映射模式,在指标间相关度较高、呈非线性变化,或数据缺漏不全等情况下仍可得到比较满意的结果,因此是一种比较理想的预测方法,具有广泛的适用范围和较高的推广价值.

Financial Crisis Warning Model based on BP Neural Network

The paper uses the BP artificial neural network to establish a model of financial crisis warning for the purpose of testing the accuracy of the methods used for financial crisis warning. The model's sample is consisted of the cross section financial indexes of 120 listed companies of which 60 companies is used as a testing sample. After repeated training and studying of the samples, we got a distinguishing-positive rates of 90.8% to the sample model setting and 90% to the testing sample.. This is a large improvement to the predicting precisions,in comparison with main-composition analysis which has got a percentage of 90 to the first sample and 81.7 to the later by using the same samples. The study indicates that BP neural network is a model without linear mapping, high in correlation among indexes, changing non-line and that a relatively satisfying outcome can be gotten even if the data are not complete or lacking. So it's a relatively optimum method used for forecasting, with wide applying area and high value of popularizing.