基于logistic回归模型的企业信用风险实证研究

发布时间：2018-09-06 08:33

【摘要】：随着我国信用市场的快速扩张,信用风险逐渐被市场参与者们所关注,信用评级机构也有着越来越高的市场地位。虽然国外信用评级行业已经走过了上百年的历程,但是在我国,这仍然是个新兴产业。国内学者一直致力于寻找适合国内企业及经济状况的信用评级方法。本文就是以此为出发点,主要介绍了logistic回归模型的相关理论,基于前人的研究,加以一定的改进,并针对某些具有相似特征的行业进行实证分析。在进行模型建立的过程中,主要有以下几个难点：一是指标的选择,影响企业信用状况的因素有很多,如何选择出风险解释性最强的一些指标是重中之重；二是违约样本与正常样本的选择,样本选择是否合适,直接影响到模型的可行性,本文不同于以往研究,采用可获得现金与本期到期债务之比来度量企业违约可能性的大小。通过显著性检验及logistic逐步选择法,最后筛选出3个指标进入logistic模型。模型对实验组判别的正确率达到了90%以上,对预测组的判断正确率为75%。且通过ROC曲线及K-S检验可以发现,模型可以有效得区分违约企业和正常企业。最后,对本文模型进行了平价并提出了自身的一些缺陷,如样本数据精度与准确性较差,模型预测能力不足等,希望以后的研究中能进一步改进。
[Abstract]:With the rapid expansion of China's credit market, credit risk is gradually concerned by market participants, credit rating agencies also have a higher and higher market position. Although the foreign credit rating industry has gone through a hundred years, but in China, this is still a new industry. Domestic scholars have been looking for credit rating methods suitable for domestic enterprises and economic conditions. This paper mainly introduces the related theory of logistic regression model, based on the previous research, to improve it, and to make empirical analysis on some industries with similar characteristics. In the process of establishing the model, there are mainly the following difficulties: first, the selection of indicators, there are many factors affecting the credit status of enterprises, how to select some of the strongest risk explanatory indicators is the most important; Second, the selection of default sample and normal sample, whether the sample selection is appropriate, directly affects the feasibility of the model. In this paper, different from previous studies, the ratio of available cash to current maturity debt is used to measure the possibility of corporate default. Through significance test and logistic stepwise selection, three indexes were selected to enter the logistic model. The accuracy rate of the model was over 90% for the experimental group and 75% for the prediction group. Through ROC curve and K-S test, it can be found that the model can effectively distinguish defaulting enterprises from normal enterprises. In the end, we put forward some defects of the model, such as the poor precision and accuracy of the sample data, the poor prediction ability of the model, and so on. We hope that further improvement can be made in the future research.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F224;F832.4

【相似文献】