中国公司债券评级方法应用研究
2015-03-26田渊博
田渊博
摘要在国内外债券评级的研究基础之上,选用MDA、Logistic模型、Probit模型以及神经网络四种债券评级方法,结合中国上市公司的风险特征,从变量甄选的角度对债券评级方法进行优化,同时采用中国上市公司数据进行实证分析.实证结论表明:甄选出的评级变量较国外常用的评级指标更好的刻画了中国上市公司的风险特征;Logistic模型、Probit模型和神经网络方法都对中国上市公司的债券有较高的评级分类能力,尤其是Probit模型和神经网络方法对中国公司债券的评级非常准确,误判率接近于0.
关键词债券评级;MDA;Logistic;Probit;神经网络;变量甄选
中图分类号 F830.91 文献标识码A
AbstractBased on the researches about bond rating at home and abroad, this paper chooses four types of methods including MDA, Logistic Model, Probit model and neural network,and according to the risk features of China list corporations, such methods were optimized by the angle of the variable selection, and the data of China list corporation was used to conduct an empirical analysis. The conclusions show that the rating variables selected can capture the risk features of China list corporations better than the rating variables often chosen in the literatures abroad, and all of Logistic Model, Probit model and neural network have much more capability of rating classification to the bonds of China list corporations, especially the rating results of the Probit model and neural network method are very precise,and the error classification rates are almost 0.
Key words bond rating; MDA; Logistic; Probit; neural network; variable selection
1引言
债券评级对于债券的发行以及债券收益率价差都有较大的影响,在债券市场的构建中具有重大的理论价值和实践意义.债券评级能够减少资本市场中的信息不对称,降低公司的融资成本,提高金融监管效率,活跃市场交易从而提高市场效率.优良的债券评级方法和技术能够为债券市场及资本市场提供准确的风险信息,提高市场资源配置的能力和效率;同时,债券评级方法及技术是债券市场和资本市场发展进步的必然产物,也是债券市场和资本市场健康发展的基础与保证.因此,债券信用评级技术的优化和提高,必然能够促进中国公司债券市场的发展与进步,以完善中国债券市场.
债券评级模型和方法的研究在国外已逾百年,从而积累了一批在理论和实践中都得……
