基于非线性插值的小企业信用评级研究

发布时间：2018-01-01 13:14

本文关键词：基于非线性插值的小企业信用评级研究　出处：《大连理工大学》2016年博士论文　论文类型：学位论文

【摘要】：小企业提供了超过80%的城镇就业岗位,创造了52.2%的税收以及58.5%的国内生产总值,其贷款需求也分别高出大型和中型企业5.9个和4.3个百分点。由于小企业自身财务信息不够公开透明,制度不够具体规范,银行较难把握其真实发展状况,导致银行不愿为小企业进行融资等原因外,商业银行缺少一套专门针对小企业贷款的信用评级体系也是其中一个关键的原因,因此亟需构建一套适用于小企业的信用评级体系。信用评级的本质是挖掘评级数据与违约风险之间的规律性联系,揭示一个客户或一笔债务的违约风险大小,评估其偿还的可能性及违约损失率大小。信用评级对违约风险识别能力的强弱直接关乎金融市场的稳定性。2008年次贷危机的发生正是源起违约风险识别错误,在此之后穆迪等权威机构的“黑箱”评级过程也备受质疑,从沉重的金融危机中可知,信用评级中指标筛选、指标赋权、信用等级确定等关键环节均要以识别违约风险为标准,否则无论多么流行、权威的信用评级体系都是不合理的。因此构建一套能够有效识别违约风险的信用评级体系是至关重要的。基于非线性插值的小企业信用评级研究,研究内容主要包括以下三部分：小企业信用评级指标体系的构建、小企业信用评分模型的建立,以及小企业信用评级模型的建立。其中,小企业信用评级体系的构建系指根据指标对违约状态的鉴别能力越大、越应保留的思路,构建一套既能反映企业客户偿还能力,又能显著区分违约状态的信用评价指标体系。小企业信用评价模型是指根据违约与违约样本的距离最大为目标函数,反推出最优信用评价方程的权重,从而建立小企业信用评分模型、进而得出小企业的信用得分。小企业信用评级模型的建立是通过非线性插值方法对旧数据进行“加权平移”,取得与“通过新、旧全部样本的统计规律挖掘出的另一套指标体系”一致的评级结果,确定出小企业的信用等级。本论文共分五章。第一章是绪论,对研究背景及意义进行了介绍,并对国内外相关研究进行了梳理；第二章是基于逻辑回归显著性判别的小企业信用评级指标体系的构建；第三章是基于“违约与非违约样本距离”最大的信用评分模型的构建；第四章是基于信用得分非线性插值的信用评级模型研究；第五章是结论及展望。本论文的主要工作及创新如下：(1)信用评级方面的工作及创新：通过旧样本的指标数据的“加权平移变换”构建信用评级模型,保证了当采用与“通过旧样本的统计规律遴选或挖掘出的一套指标体系”一模一样的指标体系进行新样本的评级时,也能得到与“通过新、旧全部样本的统计规律挖掘出的另一套指标体系”同样的评级结果.通过对旧样本数据进行加权平移变换,构建非线性插值信用评级模型,在样本增加的情况下无需重新进行“指标遴选、指标赋权、评级方程”等繁琐过程,仅仅需要把新样本的指标数据直接输入到评级方程中,便可得到与“通过新、旧全部样本的统计规律挖掘出的另一套指标体系”一致的评级结果,保证了评级指标体系的不变,事实上,任何一家评级公司的指标体系在相当长一段时期内都是不变的,而不是频繁地变动评级指标体系。并弥补了现有研究中直接根据过去样本挖掘的指标体系确定新客户信用等级,忽视加入一个或多个样本后样本的统计规律已经发生变化、旧样本挖掘的指标体系已经不适用于确定新样本的评级结果的问题。(2)指标赋权方面的工作及创新：根据“违约客户与非违约客户信用得分的组间距离越大、组内平均距离越小,则评价方程鉴别违约能力越强”的思路构建多目标规划模型,求解最优的权重系数,保证赋权后的评价得分违约鉴别能力最大。通过违约和非违约客户信用得分0的组间距离越大、组内平均距离越小,则评价得分Sj区分违约状态的能力越强的思路,设定违约非违约两类样本的组间距离与组内平均距离的比值最大为目标函数,以单一赋权的最大值和最小值为约束条件,构建线性规划模型。由于目标函数是关于信用得分Sj的函数,而信用得分Sj是关于权重wi的函数,因此目标函数是关于权重w,的函数,也就是通过求解目标规划的最优解即可得到评价指标权重wi的最优解。保证了求解的指标权重w,能够最大程度的区分违约与非违约客户的信用得分Sj,改变了现有研究中计算指标权重的过程主观性较强、无法让评价模型达到最强的违约状态鉴别能力的弊端。(3)指标遴选方面的工作及创新：通过构建评级指标与违约状态之间的逻辑回归模型,求解每个指标判别违约状态的显著性水平,遴选其中对违约状态影响显著的指标,弥补现有小企业信用评级指标体系没有根据违约状态遴选指标、无法反映指标对违约状态影响大小的不足。以违约状态yi为因变量,以评价指标xij为自变量构建逻辑回归模型,求解每个指标对违约状态判别的显著性水平,即W统计量检验概率值sigj。将概率值sigj与预先给定的显著水平α进行对比,若sigj≤α,表明第j个评价指标xij对小企业的违约状况显著影响,该指标应予以保留；反之,若sigjα,则表示第j个指标xij对小企业违约状况显著不影响,可以被剔除。保证了筛选后保留的指标能显著区分小企业的违约状态,弥补现有小企业信用评级指标体系没有根据违约状态遴选指标、无法反映指标对违约状态影响大小的不足。
[Abstract]:Small enterprises provide more than 80% of urban jobs, creating a 52.2% tax and 58.5% of GDP, the demand for loans were also higher than large and medium-sized enterprises of 5.9 and 4.3 percentage points. Due to the small enterprises financial information is not transparent, the system is not practical regulation, the bank is difficult to grasp the real development the status, cause banks' reluctance to small business financing and other reasons, commercial banks lack a specific small business loan credit rating system is one of the key reasons, the credit rating system so it is necessary to construct a set of suitable for small enterprises. The essence of credit rating and default rating data mining is the relationship of risk between, revealing a customer or a debt default risk, assess the likelihood of repayment and default loss rate. The credit rating of the default risk identification The strength of the force is directly related to the stability of financial markets.2008 the subprime mortgage crisis is the origin of the risk of default false recognition, after Moodie, the authority of the "black box" rating process has been questioned, as can be seen from the heavy financial crisis, credit rating index selection, index weight, credit rating to determine the key link to identify the risk of default as the standard, otherwise no matter how popular, the credit rating system of authority are not reasonable. So build a credit rating system can effectively identify the risk of default is very important. The research on credit rating of small enterprises based on nonlinear interpolation, the research content mainly includes the following three parts: the construction of the credit rating index system small enterprises, establish the credit scoring model for small businesses, and to establish a credit rating model for small businesses. Among them, the credit rating system for small enterprises Construction refers to the ability to identify the default state according to the index is bigger, more should be reserved for the construction of a set of ideas, which can reflect the enterprise customers the ability to repay, and can significantly distinguish default credit evaluation index system. The small enterprise credit evaluation model is based on breach of contract and breach of the sample distance as the objective function, anti the launch weight optimal credit evaluation equation, so as to establish a credit scoring model, and then draw the small business credit score. To establish a credit rating model of small enterprises is "weighted translation" of old data by nonlinear interpolation method, and the adoption of new, another set of index system of "statistical law of the whole sample of old mining the same rating results, determine the small enterprise credit rating. This paper consists of five chapters. The first chapter is the introduction, the research background and significance are introduced, and the domestic and foreign. Research carried out; the second chapter is the construction of logic regression discriminant small enterprise credit rating index system based on; the third chapter is based on the "default and non default sample from" the biggest credit scoring model construction; the fourth chapter is the research on the model of credit rating credit score based on the nonlinear interpolation; the fifth chapter is the conclusion and prospect. The main work and innovation are as follows: (1) the work and innovation of credit ratings: the old sample index data of "weighted translation" to construct a credit rating model, guaranteed when using and selection or dig out the statistics law old samples a set of index system index system the new sample rating as like as two peas, and also can get through the new, another set of index system of "statistical law of the whole sample old mined the same rating results. Weighted by the translation of the old data, construct the nonlinear interpolation model of credit rating, in the sample increases without re "index selection, index weight, rating equation and other complicated process, only need to index data of the new sample directly into the rating equation, we can get through the new, and" another set of index system of "statistical law old total samples excavated consistent rating results, the rating index system unchanged, in fact, the index system of any Rating firm are unchanged in quite a long period of time, rather than the frequent change of rating index system and make up the existing research. In the past the sample directly according to the mining index system to identify new customer credit rating, ignore to one or more of the sample after sample statistics has changed, the old mining refers to the sample Standard system is not suitable to determine the new sample rating results. (2) determine the work and innovation aspects: according to the "default and non default customer customer credit score between the groups is bigger, in the group average distance is small, the multi-objective programming model to construct the evaluation equation of differential default stronger" the idea of solving the optimal weight coefficient, ensure that the evaluation score after weighting the greatest discriminating power. By default default and non default credit score of 0 groups within the group average distance is, the smaller the distance, the evaluation score of Sj between default state more ideas, set the default non default two the ratio of the average distance between the sample group and the distance within the group as the objective function, the maximum value and the minimum value of the weighted single constraint conditions of linear programming model. The objective function is a function of the credit score of Sj However, the credit score Sj is a function of the weight of the WI, so the objective function is the weight of about W, the function is through the optimal solution can be obtained by the optimal target planning evaluation index weight solution of wi. To ensure that the w index weights, can distinguish the maximum default and non default client's credit score Sj the process of change, the subjective index weight calculation of strong, can let the evaluation model get the strongest default ability to identify defects. (3) the index selection work and Innovation: by and against the construction of evaluation index about the state of the logic regression model, the significant level of each index for default judgment. The selection of the indicators of the impact of default status significantly, make up the credit rating index system according to the existing small businesses do not have the default selection index can not reflect the index of the default state. The size of the ring. The dependent variable Yi as the default state, the evaluation index Xij as independent variables to construct a logistic regression model, for each index of the default level state identification, namely W statistic probability probability value sigj and value sigj. will be given a significant contrast, if sigj is less than or equal to alpha, that article the j index of Xij significant effect on the default status of small enterprises, the index should be retained; on the other hand, if the sigj alpha, said the j Xij index for small businesses do not affect significantly the default status, can be removed. The screen is retained after the index can significantly distinguish the default state of small enterprises, make up for the credit rating index system according to the existing small businesses do not have the default selection index can not reflect the index of the size of the state. The default problem

【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

【相似文献】