中国公司债券评级方法的应用研究

发布时间：2018-08-07 12:38

【摘要】：公司债券市场是债券市场的重要组成部分,公司债券市场的发展与完善直接关系着债券市场甚至资本市场的运行效率,因此,完善中国公司债券市场是我国资本市场健康发展的根本所在。然而,我国债券市场尤其是公司债券市场已经远远落后于股票市场；除去制度因素之外,我国公司债券市场落后的主要技术根源便是公司债券评级方法的落后。本文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,除选取部分国内外公认的财务指标外,还选取了公司控股性质,Tobin q,β以及EBIT/流动负债四个指标；同时采用中国上市公司数据对评级方法的应用能力进行实证检验,并基于评级结果,从资产定价理论出发构建出债券组合的投资策略。实证结论表明：本文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,对于训练样本,这三种债券评级方法都能够将95%以上的债券类型正确区分,尤其是Probit模型,能够将训练样本中的所有上市公司正确分类,对于测试样本,这三种评级模型均能够将93%的公司债券正确分类。综合考察训练样本和测试样本,Probit模型和BP神经网络方法的评级结果非常准确,债券评级的误判率几乎为0。
[Abstract]:The corporate bond market is an important part of the bond market. The development and perfection of the corporate bond market are directly related to the operating efficiency of the bond market and even the capital market. Therefore, improving the Chinese corporate bond market is the root of the healthy development of the capital market in China. However, the bond market, especially the corporate bond market, is far away in China. Far behind the stock market; apart from institutional factors, the main technical root of the backwardness of China's corporate bond market is the backwardness of corporate bond rating methods.
Based on the study of bond rating at home and abroad, this paper selects four bond rating methods, MDA, Logistic model, Probit model and neural network. It combines the risk characteristics of Chinese listed companies and optimizes the bond rating method from the angle of variable selection. Holding nature, Tobin Q, beta and EBIT/ mobile liabilities four indicators, and using the data of Chinese listed companies to test the application capacity of the rating method, and based on the rating results, the investment strategy of the bond portfolio is constructed from the asset pricing theory. The empirical conclusion shows that the rating variables selected in this paper are more commonly used than the foreign countries. The rating indicators better depict the risk characteristics of Chinese listed companies; the Logistic model, Probit model and neural network approach have higher rating classification ability for Chinese listed companies. For training samples, these three bond rating methods can correctly distinguish over 95% of the bond types, especially the Probit model, All listed companies in the training sample can be correctly classified. For the test samples, the three rating models can correctly classify 93% of the corporate bonds. The comprehensive inspection of training samples and test samples, the Probit model and the BP neural network method are very accurate, and the rate of miscarriage of debt vouchers is almost 0..
【学位授予单位】：东北财经大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F832.51;F224

【参考文献】