基于信息敏感性的小企业信用评级模型研究

发布时间：2018-04-01 18:28

  本文选题：信用评级　切入点：信息敏感性　出处：《大连理工大学》2016年博士论文

【摘要】：作为国民经济和社会发展重要组成部分的小企业,信用评级的结果直接影响其融资的渠道、可能性及成本。甚至,信用评级结果会影响小企业购销合同的签订、招投标的参加及政府采购的争取等诸多方面。特别是多年以来小企业贷款贵和贷款难的问题始终都未得到有效解决。这一问题的解决固然复杂困难,但鉴于目前许多商业银行等金融机构已将对贷款企业客户的信用评级结果纳入到信贷审批程序之中,显然若能科学合理地对小企业进行信用评级必将有助于贷款贵和贷款难问题的解决。此外,信用评级也事关着国家金融体系的安全。因此,开展小企业信用评级的研究至关重要。小企业信用评级的本质在于其信用风险的评价。为了通俗形象地反映企业信用风险的相对大小,通常会将企业的信用风险进行等级划分。而信用评级指标体系和信用评级指标的赋权是信用风险评价的基础。其中,信用评级指标体系是通过信用评级指标筛选,即通过剔除信用风险识别能力弱和反映信息重叠程度高的信用评级海选指标实现的；信用评级指标赋权是指确定信用评级指标的权重。本文主要聚焦于小企业信用风险的评级及其合理性的检验,主要工作如下：(1)为了选取信用风险识别能力大的信用评级指标,提出了一种基于信用风险识别能力的信用评级指标遴选方法。首先,将信用评级指标的取值分为违约与非违约两组,根据一个信用评级指标取值的组间变异与组内变异的相对差距越大,该信用评级指标对违约与否的影响越显著,该信用评级指标越应予以保留的F检验思路,剔除对信用违约与否影响不显著的少量信用评级指标,以保证保留下来的信用评级指标皆对信用违约状态有一定的影响。之后,通过主成分占待筛信用评级指标集信息、的比例与每个被保留的主成分对信用评级指标偏导数的乘积的和,表示待筛信用评级指标集信息受该指标大小变化影响的敏感程度,称之为指标的信息、敏感性。在此基础上,根据信用评级指标的信息敏感性越大,该指标解释待筛信用评级指标集蕴含的信用风险信息的能力越大,该指标识别信用违约综合风险的能力也越大,该指标越应予以保留的信息敏感性思路,进一步地剔除信用违约综合风险识别能力弱的信用评级指标。该方法克服了现有研究仅依赖信用评级指标对违约与否的影响选取信用风险识别能力大的指标的不足,也克服了现有主成分降维方法遴选指标时仅依据指标的单个因子载荷无法合理筛选指标的不足。(2)为了降低信用评级指标彼此之间的信息重叠程度,提出了一种基于信息重叠双重剔除的信用评级指标遴选方法。首先,该方法通过一组信用评级指标的病态指数确定一组信用评级指标整体的信息重叠程度,并以一个评级指标被剔除后该组信用评级指标病态指数减少的幅度确定该指标与指标集内其余指标间的信息重叠程度,简称为该信用评级指标的信息重叠程度。通过剔除信息重叠程度高的指标,实现信用评级指标间整体信息重叠程度的快速降低。之后,为了避免指标体系整体信息重叠程度低而部分指标间信息重叠高,进一步地剔除相关程度高的任两个指标中信息敏感性小的指标。该方法克服了现有研究仅仅考虑两两指标间的信息重叠的降低而不考虑评级指标间整体信息重叠程度的不足。(3)为了确定小企业信用评级指标的权重,提出了一种基于双重信用违约风险识别的信用评级指标赋权方法。首先,该方法根据信用评级指标识别信用违约与否提供的信息量(信息增益)越大,该信用评级指标识别违约与否的能力越大,该指标的权重理应越大的信息增益赋权思路,实现信用评级指标的第一重赋权。之后,鉴于信息敏感性能够反映信用评级指标识别信用违约综合风险能力的大小,根据信用评级指标的信息敏感性越大、该信用评级指标信用违约综合风险识别的能力也越大、该指标的权重理应也越大的信息敏感性赋权思路,实现信用评级指标的第二重赋权。第一重赋权虽然利用了违约与否所蕴含的信用风险的价值,但违约与否毕竟不能精细地刻画信用风险。第二重赋权虽然体现了信用评级指标识别信用违约综合风险的能力,但却没有有效利用违约与否这个宝贵的反映信用风险的指标。有鉴于此,最后通过乘法集成归一法对上述两种赋权方法确定的信用评级指标的权重进行了综合,实现了两种赋权方法的取长补短,克服了现有研究在对信用评级指标赋权时仅考虑对违约状态的影响程度或不考虑对违约状态的影响程度的不足。同时,基于信用评级指标的遴选和赋权确定了小企业的信用综合得分,并据此划分了不同信用等级对应信用得分区间,进而实现了小企业信用等级的确定。(4)为了检验小企业信用评级指标体系的合理性,根据指标体系对违约与否判别的准确率越高,指标体系整体上识别基本信用风险的能力越大的思路,提出了基于贝叶斯判别的信用评级指标体系合理性检验模型,克服了现有研究完全不考虑指标在单位及量纲上的差异及错误使用多重决定系数的不足。此外,为了检验小企业信用评级指标赋权的合理性,根据全部样本企业间信用综合得分与实际违约损失率间的不一致,提出了基于信用非一致性比率的信用评级指标赋权合理性检验方法,克服了现有研究缺乏此类客观检验方法的不足。并在实证研究中,从对违约状态判别的准确率和信息重叠程度两个不同角度,将本文构建的小企业信用评级指标体系与现有研究的两套小企业信用评级指标体系进行了定量对比,进一步验证了利用本文方法构建信用评级指标体系的合理性。
[Abstract]:As an important part of the national economic and social development of small enterprises, credit rating results directly affect its financing channels, and the possibility of cost. Even, the credit rating results will affect the purchase contract is signed for small businesses, and to participate in the bidding and government procurement bidding for many years. Especially for small business loans and expensive loan difficult problem has not been solved effectively. To solve this problem is difficult, but because of the many commercial banks and other financial institutions have credit rating results of the loan business customers into the credit approval process, obviously if reasonably on the small business credit rating will help loan and your loan is difficult to solve the problem. In addition, the credit rating is also a matter of national security of the financial system. Therefore, it is important to study development of small enterprise credit rating of small enterprises. The essence of the credit rating industry is to evaluate the credit risk. In order to reflect the relative size of the popular image of enterprise credit risk, usually will rank the credit risk of the enterprise. And the weighted index system of credit rating and credit rating is the basis of credit risk evaluation. The credit rating index system by credit rating index selection, through the recognition of credit risk elimination ability and weak implementation of the credit rating index reflects the high degree of overlap of audition information; credit rating index weight refers to the weight to determine the credit rating index. This paper mainly focused on small business credit risk rating and the rationality of the test, the main work is as follows: (1) to the credit rating index the selection of the credit risk recognition ability, puts forward a selection method of credit risk recognition ability of the credit rating index. Based on the first letter With the value of rating for breach and non breach two groups, according to the relative gap between a credit rating index value of the variance between groups and within group variation is bigger, the credit rating index for breach of contract and whether the influence is more significant, the credit rating index should be retained more F test ideas, excluding significant the effect of credit default and not a credit rating index, to ensure that the retained credit rating index have a certain influence on the credit default. Then, by principal components accounted for the screen credit rating index set of product information, and the proportion of each component is reserved for the credit rating index and partial derivative said to be more sensitive, screen credit rating index set information by the index size change, called index information sensitivity. On this basis, according to the information sensitivity of credit rating index is the index. The ability to explain the credit risk information to screen the credit rating index set contains a larger capacity of the identification index of credit default risk is more comprehensive, more sensitive information ideas of the index should be retained, further eliminate credit default risk recognition ability of the comprehensive credit rating index. This method overcomes the shortcomings of existing research only rely on the lack of credit rating index of default or not the influence of selection of large index of credit risk identification ability, and overcomes the problem of principal components method selection index only based on the deficiency of single factor load index cannot reasonable screening index. (2) in order to reduce the degree of information overlap between each of the credit rating index, is proposed a method of selecting information overlapping of the credit rating index based on double elimination. Firstly, the method to determine a set of credit through a set of credit rating index morbid index The overall rating index information overlap degree, and with a rating index were eliminated after the group credit rating index index to reduce the magnitude of morbid information to determine the degree of overlap index and other indexes between the index set, referred to as information overlap degree of the credit rating index. Through eliminating a high degree of overlap of information index, fast to reduce the implementation of credit rating index between the overall information overlap degree. Then, in order to avoid the whole information index system of low degree of overlap and part of the overlapped information index is high, further eliminating a high degree of correlation as the two indicators in the information sensitivity index. The existing small studies considered only 22 index information overlap between the reduced and not considering the lack of information between the overall rating index of the degree of overlap to overcome this method. (3) to determine the weight of the credit rating index of small enterprises, propose a method based on double Credit default risk recognition credit rating index weighting method. First, according to the credit rating index of credit default or identification information (information gain) greater recognition ability of the credit rating index of default or not is greater, the weight of the index should be more information gain weighting ideas, realization of credit the first heavy weight rating index. Then, in view of information sensitivity can reflect the credit rating index of credit default risk identification capability, according to the information sensitivity of credit rating index is bigger, the credit rating index of credit default risk identification is greater, the weight of the index should be more information sensitivity weighting ideas the credit rating index, the second weighting. The first weighting although use default or not contained in the value of credit risk, but the default or not after all Not fine description of credit risk. The second weighting reflects the credit rating index of credit default risk recognition ability, but there is no effective use of this valuable default or not reflect the credit risk indicators. In view of this, the integrated weight multiplication normalization method for the two kinds of weighting method to determine the credit rating index of overall, realized two kinds of weighting methods complement each other, to overcome the existing research on the weight of the credit rating index when only considering the influence degree of the default state or not considering the insufficient degree of influence on the default state. At the same time, the credit rating index selection and weight to determine the comprehensive score based on small business credit, and accordingly division of the different credit rating corresponds to the credit score interval, thus confirming the credit rating of small enterprises. (4) in order to test the small enterprise credit rating means The rationality of index system, according to the index system of breach of contract and the accuracy of the discriminant is higher, the overall index system of thinking ability to identify the basic credit risk more, proposed Bayesian credit rating index system to verify the rationality of model based on the existing research, to overcome the lack of index without considering the differences in unit and dimension the use of multiple coefficient of determination and error. In addition, in order to test the rationality of weighted credit rating index of small enterprises, according to the comprehensive score of all samples between the enterprise and the actual credit default loss rate is not consistent between the proposed reasonable test method of weighting non consistency ratio of credit based on credit rating index, overcomes the shortcomings of the existing lack of such the objective test method. And in the empirical study, from the default state estimation accuracy and information overlap of two different angles, the The small business credit rating index system is quantitatively compared with the existing two sets of small business credit rating index system, further validates the rationality of using this method to build credit rating index system.

【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F276.3;F832.4
，

本文编号：1696743