我国商业银行信用风险识别的多模型比较研究.docVIP

我国商业银行信用风险识别的多模型比 较研究 刘祥东王未卿 北京科技大学东凌经济管理学院 笔者以我国A股325家上市公司2011年和2012年的财务数据作为样本，利用贝 叶斯判别法、Logistic回归模型和BP神经网络模型对信用风险进行识别，进而 比较三类模型的准确性、预测能力和稳定性，发现三类模型对信用风险识别的准 确率依次增高，但仍然都存在较大的概率将信用状况非健康公司识别为健康公司; W叶斯判别法和Logistic冋归模型识别出的重要财务指标能够有效解释公司的 信用状况，而BP神经网络模型则缺乏对识别结果的解释能力。比较结果对商业银 行选择和使用合适的信用风险识别技术具有重要的参考价值。 关键词： 災叶斯判别法;Logistic冋归模型;BP神经网络模型;信用风险识别; 基金：国家自然科学基金项0 （71420107023） A Comparative Study on Three Models for Credit Risk Identification in Chinese Commercial Banks LIU Xiang-dong WANG Wei-qing Donlinks School of Economics and Management，University of Science and Technology; Abstract： This paper takes 325 listed companies of Chinese A-shares and their 2011 and 2012 financial data as testing samples, it employs the tests of normality, significance and correlation to choose effective credit risk identification indexes, and respectively uses the Bayes discriminant method, Logistic regression model and BP neural network model to distinguish credit risk, and then compares their accuracy, predictability and stability. The results show that the degrees of accuracy for identifying credit risks are gradually increasing by the above three models. However, there is still the probability that the company whose credit situation is unhealthy can be identified as healthy one. The important financial indicators identified by Bayes discriminant method and Logistic regression model can effectively explain the company’s credit status, while BP neural network model is unable to interpret the recognition results. The results provide an important reference for selecting and using proper technology of credit risk identification in commercial banks. Keyword： Bayes Discriminant Method; Logistic Regression Model; BP Neural Network Model; Credit Risk Identification; 一、学术综述 目前，流行的信用风险识别方法主要可以分为三类：第一类是偏重定性分析的模 糊判别法；第二类是偏重计量分析的定量方法；第三类是采用现代计算技术的 高级人工智能方法。在模糊判别法方面，周春喜等（ 2004）运用模糊数学理论对 银行信用风险进行综合评价；刘澄等（2012）釆用可信性理论识别和度量商业 银行对中小企业的信贷风险。这些方法虽然加深了人们对信用风险构成要素的理 解，但由于更偏重于定性的方式,个人的主观判断对最终结果产生较大影响，因而 并未发展成为信用