WO2017181346A1 - 基于信用相似度最大的信用等级最优划分方法 - Google Patents
基于信用相似度最大的信用等级最优划分方法 Download PDFInfo
- Publication number
- WO2017181346A1 WO2017181346A1 PCT/CN2016/079681 CN2016079681W WO2017181346A1 WO 2017181346 A1 WO2017181346 A1 WO 2017181346A1 CN 2016079681 W CN2016079681 W CN 2016079681W WO 2017181346 A1 WO2017181346 A1 WO 2017181346A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- credit
- credit rating
- customer
- rating
- lgd
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
Definitions
- the invention relates to an optimal credit rating method based on credit similarity, and belongs to the technical field of credit services.
- the essence of credit rating is to classify customers according to their credit status and reveal different credit risk levels of different levels of customers.
- the division of credit ratings is the final result of the credit rating. If the credit rating is unreasonable, it will mislead investors and the public to make wrong investment decisions. Therefore, it includes the selection of credit rating indicators, the empowerment of indicators, the determination of customer credit risk evaluation scores, and the division of credit ratings. In the credit rating system, the division of credit ratings is particularly important.
- the first type of existing credit rating is divided into credits by the credit rating, or according to the idea that the default probability is greater than a certain threshold.
- the patent number of the Intellectual Property Office of the People's Republic 200810139934.8's 'Credit Rating Management consulting System' includes financial analysis, credit rating, risk management system, etc. 15 Modules with clear structure, easy to expand, easy to reuse and so on.
- 'Credit Rating System' of Patent No. 201010546434.3 of the State Intellectual Property Office of the People's Republic of China It provides an informationization system for credit rating service organizations to carry out credit rating business.
- the first type of credit rating related patents have the following deficiencies: the classified credit rating does not satisfy 'the higher the credit rating, the lower the default loss rate' This credit nature attribute. It leads to many rating systems that look perfect on the indicators. When evaluating customers, they often get customers with high credit ratings, and the corresponding default loss rate is not low.
- the second type of existing credit rating is mainly divided by 'the higher the credit rating, the lower the default loss rate'
- the default pyramid standard is used to classify the credit level to which the customer belongs.
- the patent number of the Intellectual Property Office of the People's Republic of China is 201210201461.6 Credit rating system and method based on credit rating and default loss rate'
- the patent number of the Intellectual Property Office of the People's Republic of China is 201210201114.3 Credit rating adjustment method based on credit rating and default loss rate', these two patents are based on 'the higher the credit rating, the lower the default loss rate'
- the standard divides the credit rating, and the higher the credit rating, the lower the default loss rate corresponding to the loan customer.
- the second type of invention patents are graded according to the default pyramid standard with higher credit rating and lower default loss rate, which satisfies the essential characteristics of credit rating.
- these two patents do not consider the fact that the greater the similarity of credit, the more should be divided into the same credit rating, which will lead to the logic of different credit ratings. confusion.
- the invention adopts the minimum deviation within the group of the credit risk evaluation scores in the same level, and the maximum deviation between the groups of the credit risk evaluation scores in different levels as the objective function, and strictly increases the default loss rate of the credit level from high to low as the constraint.
- the multi-objective planning model divides the optimal credit rating, and ensures that customers with different credit status are divided into different levels under the premise that customers with similar credit status are divided into the same level, and at the same time, the credit rating result can be guaranteed to meet the credit rating.
- the default the lower the default loss rate, the default pyramid standard.
- the object of the present invention is to provide a customer whose credit status is similar to the same level under the premise of satisfying the credit attribute with higher credit rating and lower default loss rate, and customers with different credit status are divided into different levels.
- the optimal division method of credit rating is to provide a customer whose credit status is similar to the same level under the premise of satisfying the credit attribute with higher credit rating and lower default loss rate, and customers with different credit status are divided into different levels. The optimal division method of credit rating.
- the default loss function of the credit rating from high to low is strictly increased.
- the multi-objective programming model divides the optimal credit rating.
- Step 1 Determine the credit risk assessment score S i
- Empowerment credit risk assessment indicators steps through the mean square error method 1) Obtaining the weight of indicators in the credit risk evaluation index system, the greater the mean square error of the indicators, the greater the weight;
- Constructing the credit risk evaluation index system and determining the index weight are the basis for calculating the credit risk assessment score S i and have several determination methods.
- Step 2 Data Import
- Step 3 Credit rating
- the credit rating optimization algorithm based on the greatest credit similarity includes:
- Constraint 1 The credit loss rate of each credit rating increases from high to low.
- LGD 1 0 ⁇ LGD 1 ⁇ LGD 2 ⁇ LGD 3 ⁇ LGD 4 ⁇ LGD 5 ⁇ LGD 6 ⁇ LGD 7 ⁇ LGD 8 ⁇ LGD 9 ⁇ 1 .
- the multi-objective programming model is composed of multi-objective programming model to obtain the optimal credit rating.
- the credit rating results are guaranteed to be similar under the default pyramid criteria that meet the higher credit rating and lower default loss rate. Customers are divided into the same level, and customers with large differences in credit status are divided into different levels.
- the present invention provides an optimal credit classification method based on the greatest credit similarity, so that the credit rating result is satisfied.
- customers with similar credit status are classified into the same level, and customers with large differences in credit status are classified into different levels. Taking the data of all rural household loans in each province of China as a sample and the loan data of all small enterprises in a regional commercial bank in China as samples, they all met the satisfaction of 'the higher the credit rating and the lower the default loss rate'. Under the condition of the essential attribute of this credit rating, customers with similar credit status are classified into the same level, and customers with large differences in credit status are classified into different levels of credit classification results.
- the second is to avoid infinite adjustments to get 'the higher the credit rating, the lower the default loss rate' The result of the adjustment; because the credit rating results are adjusted according to common sense, due to the adjustment of the number of customers or the default rate of a credit rating, it is bound to cause changes in the number of neighboring customers and the default rate, because the rational number between any two points is infinite common sense. It is simply impossible to make a reasonable credit rating by artificial adjustment.
- the third is to ensure that the credit rating results have the advantage of the stability of the score interval, and avoid the length of the score interval is too large or too small. If the length of the credit risk assessment score interval is too small, the customer's credit risk assessment score will change slightly, the customer's credit rating will also change, and the corresponding customer's default loss rate will also change. If the length of the credit risk assessment score interval is too large, the customer's credit risk assessment score will change greatly, and the customer's credit rating will not change. Therefore, if the credit rating interval does not have the advantage of stability, it will be misleading to the loan pricing or investment decision.
- the fifth is the credit rating division result of this method, which not only gives the credit rating of the customer's liquidity as in the existing research and practice, but also gives the default loss rate of each credit rating, compared with the existing bank credit rating system. It reveals more information that the public needs to know more about.
- Figure 1 is a pyramidal diagram of the distribution of credit ratings and default loss rates.
- Figure 2 is a distribution map of credit rating and default loss rate.
- AAA, AA, A, BBB, BB, B, CCC, CC, C represent the 9 levels of credit rating from high to low, and the length of the horizontal line inside the pyramid triangle represents the default loss rate of the corresponding level
- the invention discloses a workflow based on an optimal credit rating method with the greatest credit similarity.
- the invention divides the credit rating according to the criterion of the greatest credit similarity, realizes the function of dividing the credit grade, and ensures that the credit condition is similar under the premise that the credit result is higher in the credit rating and the lower the default loss rate.
- the customers are divided into the same level, and customers with large differences in credit status are divided into different levels of credit rating.
- a regional commercial bank in China is located in 1814 of 28 cities including Beijing, Tianjin and Shanghai.
- the pen industry small business loan data is an empirical sample, and the empirical analysis of the solution of the invention is as follows:
- Step 1 Determine the credit risk assessment score S i
- Empowerment credit risk assessment indicators steps through the mean square error method (1) Obtaining the weight of indicators in the credit risk evaluation index system, the greater the mean square error of the indicators, the greater the weight;
- the credit risk assessment indicator system is shown in column 2 of Table 1, and the indicator weights are shown in column 3 of Table 1.
- Constructing the credit risk evaluation index system and determining the index weight are the basis for calculating the credit risk assessment score S i and have several determination methods.
- Step 2 Data Import
- the credit risk assessment score S i of the 1814 loan customers, the receivable unpaid principal and interest L ki , and the receivable principal and interest R ki data are imported into the Excel file, and are ranked in descending order of the credit risk assessment score S i from high to low.
- Step 3 Credit rating
- the credit rating division optimization algorithm based on the credit similarity is used to classify the credit rating of the customer, and the division result is directly displayed on the Excel interface.
- the credit risk assessment score interval, the objective function value, and the pyramid distribution of each credit rating default loss rate are shown in Figure 1.
- the credit rating optimization algorithm based on the greatest credit similarity includes:
- the objective function 1 ensures that the closer the credit risk assessment scores are, the more the customers can be divided into the same credit rating, and the customers who have different credit risk assessment scores are divided into the same credit rating, causing the credit risk assessment score interval length to be too large, resulting in the credit rating interval. There is no distinction.
- the objective function 2 ensures that the credit risk assessment score difference between different credit ratings is as large as possible, avoiding the credit risk evaluation score interval being too small and causing the customer's credit status to change slightly, and the credit rating will also be too sensitive and stable. Insufficient drawbacks.
- Constraint 1 The credit loss rate from high to low default rate is strictly increased, ie 0 ⁇ LGD 1 ⁇ LGD 2 ⁇
- LGD 3 ⁇ LGD 4 ⁇ LGD 5 ⁇ LGD 6 ⁇ LGD 7 ⁇ LGD 8 ⁇ LGD 9 ⁇ 1 .
- Constraint 1 guarantees that the credit rating result satisfies the credit attribute of 'higher credit rating and lower default loss rate' by restricting the credit rating from high to low default loss rate, which changes the existing rating system. The resulting credit rating is high, but the default loss rate is not low.
- Constraint 2 calculates the default loss rate of each credit rating loan customer by comparing the loan customer's receivable principal and interest L ki with the receivable principal and interest R ik . It ensures that the calculation of the default loss rate can truly reflect the bank's losses.
- the objective function based on the credit risk assessment score in the present invention is not used, only the division method of the existing patents (patent numbers: 201210201461.6 and 2012102001114.3) will result in the credit rating result not having the stability of the score interval.
- the advantage is that the length of the credit risk assessment score interval is different, too large or too small. If the length of the credit risk assessment score interval is too small, the customer's credit risk assessment score will change slightly, and the customer's credit rating will also change, that is, the credit risk assessment score interval is too sensitive. If the length of the credit risk assessment score interval is too large, the customer's credit risk assessment score will change greatly, and the customer's credit rating will not change, that is, the credit risk assessment score interval does not have discrimination.
- a regional commercial bank in China is located in 1814 of 28 cities including Beijing, Tianjin and Shanghai.
- the pen industry small business loan data is an empirical sample, and the credit classification according to the hierarchical division method of the present invention is used, and the credit rating result based on the greatest credit similarity is obtained, as shown in Table 2.
- the third column of Table 2 is the credit risk assessment score interval for each credit rating
- the fourth column of Table 2 is based on the third The length of the credit risk assessment score interval determined by the column.
- the minimum length of the scoring interval is 1.05, which is the mean value of the difference between the credit risk assessment scores of two of the 1814 loan customers. Times, so the credit risk assessment score interval has a certain degree of discrimination.
- the default loss rate LGD k in column 5 of Table 2 is the horizontal axis
- the credit level k in the second column of Table 2 is the vertical axis
- the pyramid distribution map of the default loss rate of each credit rating is obtained, as shown in Fig. 1.
- Figure 1 satisfies the default pyramid standard with higher credit rating and lower default loss rate, which can be seen from the different levels of default loss rate in column 5 of Table 2.
- the present invention has various specific embodiments, and the method for optimally classifying credit ratings based on the greatest credit similarity is adopted in the present invention. All technical solutions formed by equivalent replacement, or equivalent transformation, fall within the scope of the claimed invention.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Business, Economics & Management (AREA)
- Technology Law (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
一种基于信用相似度最大的信用等级最优划分方法,属于信用服务技术领域。提供一种在满足信用等级越高、违约损失率越低的信用本质属性的前提下,保证信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同等级的信用等级最优划分方法。通过建立以同一等级中信用风险评价得分的组内离差最小、不同等级中信用风险评价得分的组间离差最大为目标函数,以信用等级由高到低的违约损失率严格递增为约束的多目标规划模型划分最优的信用等级。
Description
技术领域
本发明涉及一种基于信用相似度最大的信用等级最优划分方法,属于信用服务技术领域。
背景技术
信用评级对当代社会有极其重要的影响。不论是主权信用评级、企业信用评级、银行信用评级,还是个人信用评级,若信用等级划分不合理,必将误导债权人和社会公众。信用评级结果的变动直接反映经济状态的变化,引起投资人和社会公众的密切关注。主权信用评级结果的变化反映一国经济状况的变化,公司债券评级结果的变化又标志着工商企业或金融企业经营状况的变化。
信用评级的本质是根据客户信用状况对客户进行等级划分,揭示不同等级客户的不同信用风险水平。信用等级的划分是信用评级的最终结果。若信用等级划分不合理,会误导投资者和社会公众做出错误的投资决策,因此在包括信用评级指标的遴选、指标的赋权、客户信用风险评价得分的确定以及信用等级的划分等步骤构成的信用评级体系中,信用等级的划分尤其重要。
第一类现有信用等级划分主要通过信用评价得分属于某个区间,或根据违约概率大于特定阈值的思路来划分客户所属的信用等级。中华人民共和国知识产权局专利号为
200810139934.8 的 ' 征信评级管理咨询系统 ' 包括财务分析、信用评级、风险管理系统等 15
个模块,具有结构清晰,易于扩展,易于重用等特点。中华人民共和国国家知识产权局专利号为 201010546434.3 的 ' 信用评级系统 '
为信用评级服务机构提供了一种开展信用评级业务的信息化系统。美国专利商标局专利号为 6965878 的 'Currency and credit rating
system for business-to-business transaction'
通过信用得分属于不同区间的方法来划分信用等级。世界知识产权组织专利号为 WO/2012/012623 的 'CREDIT RISK MINING'
利用企业财务、宏观环境等多种数据,提供了企业信用等级变化的概率测算、违约率测算等模型。
第一类信用评级相关专利存在以下不足:划分的信用等级不满足 ' 信用等级越高、违约损失率越低 '
这一信用本质属性。导致很多在指标上看上去很完美的评级体系,对客户评价时往往得到信用等级很高的客户、对应违约损失率反而不低的怪现象。
第二类现有信用等级划分主要通过 ' 信用等级越高、违约损失率越低 '
的违约金字塔标准来划分客户所属的信用等级。中华人民共和国知识产权局专利号为 201210201461.6 的 '
基于信用等级与违约损失率匹配的信用评级系统与方法 ' ,以及中华人民共和国知识产权局专利号为 201210201114.3 的 '
基于信用等级与违约损失率匹配的信用评级调整方法 ' ,这两项专利根据 ' 信用等级越高、违约损失率越低 '
的标准划分信用等级,保证信用等级越高的贷款客户对应的违约损失率越低。
第二类两项发明专利根据信用等级越高、违约损失率越低的违约金字塔标准进行等级划分,满足信用评级的本质特征。但由于研究问题的角度不同,这两项专利并没有考虑信用相似度越大、越应划分为同一信用等级的准则,会导致信用风险评价得分相同的不同客户却被划分为不同信用等级的逻辑混乱。
本发明通过以同一等级中信用风险评价得分的组内离差最小、不同等级中信用风险评价得分的组间离差最大为目标函数,以信用等级由高到低的违约损失率严格递增为约束的多目标规划模型划分最优的信用等级,在信用状况相似的客户划分为同一个等级的前提下,确保信用状况不同的客户划分为不同的等级,同时能保证信用等级划分结果满足信用等级越低、违约损失率越高的违约金字塔标准。
发明内容
本发明的目的是提供一种在满足信用等级越高、违约损失率越低的信用本质属性的前提下,保证信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同等级的信用等级最优划分方法。
本发明的技术方案:
通过建立以同一等级中信用风险评价得分的组内离差最小、不同等级中信用风险评价得分的组间离差最大为目标函数,以信用等级由高到低的违约损失率严格递增为约束的多目标规划模型划分最优的信用等级。
一种基于信用相似度最大的信用等级最优划分方法,步骤如下 :
构建信用风险评级体系,包括建立信用风险评价指标体系、信用风险评价指标的赋权、建立客户信用风险评价方程以及划分客户信用等级;依次通过建立信用风险评价指标体系、信用风险评价指标的赋权和建立客户信用风险评价方程确定第
i 个客户的信用风险评价得分 S
i ,为信用等级的划分提供数据基础,通过信用风险评价得分
S
i 将 客户划分为 9 个信用等级;其中, i =1, 2, …n ,
n 为待划分信用等级的全部客户数目。
步骤 1 :确定信用风险评价得分 S
i
1) 建立信用风险评价指标体系:首先通过 Fisher
判别的方法在海选指标中遴选显著区分违约与非违约客户的指标;再通过相关分析法从上述显著区分违约与非违约客户的指标中删除反映信息重复的指标,得到信用风险评价指标体系;
2) 赋权信用风险评价指标:通过均方差方法对步骤 1)
得到信用风险评价指标体系中指标赋权,指标的均方差越大,权重越大;
3)
建立客户信用风险评价方程:对信用风险评价指标体系中指标与指标的权重进行线性加权,建立客户信用风险评价方程
S
i =
∑ω
j
x
ij , 确定第 i
个客户的信用风险评价得分 S
i ;其中, ω
j 第
j 个指标的权重, x
ij 第 j 个指标下第 i 个客户的数值,
i =1, 2, …n , j =1, 2, …m , n
为待划分信用等级的全部客户数目, m 为信用风险评价指标体系中指标的个数。
构建信用风险评价指标体系及确定指标权重是计算信用风险评价得分
S
i 的基础,并且具有若干种确定方法。
步骤 2 :数据导入
将步骤 1 中得到待划分的所有客户的信用风险评价得分
S
i 、应收未收本息 L
ki 、应收本息
R
ki 的源数据导入到 Excel 文件中,按照信用风险评价得分从高到低降序排列;
步骤 3 :信用等级划分
利用基于信用相似度最大的信用等级划分优化算法,对客户进行信用等级划分,并将划分结果直接显示于 Excel
界面;
基于信用相似度最大的信用等级划分优化算法包括:
(1) 目标函数 1 :以同一等级中客户的信用风险评价得分组内离差最小,即: min f
1 =g
1(S
k ,S
ki ) ,其中,
S
k 表示第 k 个信用等级中所有客户信用风险评价得分的均值,
S
ki 表示第 k 个信用等级中第 i 个客户的信用风险评价得分, k
=1, 2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, … 。
目标函数 2 :以不同信用等级中客户信用风险评价得分的组间离差最大,即: max f
2=g 2(S
k ,S)
,其中, S
k 表示第 k 个信用等级中所有客户信用风险评价得分的均值, S
表示 9 个信用等级中所有客户信用风险评价得分的均值, k =1, 2, 3, 4, 5, 6, 7, 8, 9 。
(2) 约束条件 1 :各信用等级从高到低违约损失率严格递增
即 0<LGD 1<LGD
2<LGD 3<LGD
4<LGD 5<LGD
6<LGD 7<LGD
8<LGD 9 ≤ 1 。
约束条件 2 :第 k 个信用等级违约损失率
LGD
k 计算的等式约束,即 LGD
k
=h( L
ki , R
ki )
;其中, L
ki 表示第 k 个 信用 等级中第 i
个客户的应收未收本息, R
ki 表示第 k 个 信用 等级中第 i
个客户的应收本息, k =1, 2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, … 。
通过确定步骤 3 所述目标函数 1 、目标函数 2 、约束条件 1 和约束条件 2
组成的多目标规划模型,通过多目标规划模型的求解得到最优的信用等级划分,使得信用等级划分结果在满足信用等级越高、违约损失率越低的违约金字塔标准下,保证信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同的等级。
本发明的有益效果:
一是本发明提供了一种基于信用相似度最大的信用等级最优划分方法,使得信用等级划分结果在满足 '
信用等级越高、违约损失率越低 ' 这一信用评级的本质属性条件下,确保 信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同的等级。
以中国某大型商业银行各省的全部农户贷款数据为样本和以中国某区域性商业银行的全部小企业的贷款数据为样本,均得到了在满足 ' 信用等级越高、违约损失率越低 '
这一信用评级的本质属性条件下,确保 信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同的等级的信用等级划分结果。
二是避免了无穷多次调整方可得到 ' 信用等级越高、违约损失率越低 '
的调整结果;因为按照常理来调整信用评级结果,由于一个信用等级客户数量或违约率的调整、势必引起相邻等级客户数量和违约率的变化,由于任意两点间的有理数是无穷多的常识,经过人为调整满足合理的信用等级划分简直是不可能的。
三是保证信用等级划分结果具有评分区间稳定性的优点,避免评分区间长度过大或过小。如果信用风险评价得分区间长度过小,客户的信用风险评价得分稍微变化,客户的信用等级也会发生变化,相应的客户的违约损失率也发生变化。如果信用风险评价得分区间长度过大,客户的信用风险评价得分发生较大变化,客户的信用等级也不会发生变化。因此,信用等级划分区间如果不具有稳定性的优点必将对贷款定价或者投资决策造成误导。
四是根据不同信用等级的违约状况可以对贷款和债券等金融工具进行弥补违约风险溢酬的合理定价。
五是本方法的信用评级划分结果,不但像现有研究和实践那样给出了客户清偿能力的信用等级排序,而且给出了每一个信用等级的违约损失率,比现有的银行信用评级系统揭示了更多、公众更需要了解的信息。
六是根据信用评级结果揭示的不同等级的违约率、可使商业银行、债券投资者等债权人和社会公众了解每一个信用等级的违约状况,进行投资决策。
附图说明
图 1 是信用等级与违约损失率分布金字塔图。
图 2 是信用等级与违约损失率不匹配的分布图。
图中: AAA 、 AA 、 A 、 BBB 、 BB 、 B 、 CCC 、 CC 、 C
代表信用等级由高到低的 9 个等级,金字塔三角形内部的横线的长度代表相应等级的违约损失率,图 1 中 9 个等级的违约损失率 LGD 满足:
LGDAAA=0.130% , LGDAA=0.263% , LGDA=0.684% ,
LGDBBB=6.040% , LGDBB=9.543% , LGDB=24.452% ,
LGDCCC=33.868% , LGDCC=35.448% , LGDC=90.044%
;图 2 中信用等级低的 CCC 等级的违约损失率小于信用等级高的 B 等级的违约损失率。
具体实施方式
以下结合附图和技术方案,进一步说明本发明的具体实施方式。
本发明揭示了一种基于信用相似度最大的信用等级最优划分方法的工作流程。
本发明通过基于信用相似度最大的准则对信用等级进行划分,实现信用等级的划分功能,使划分结果在满足信用等级越高、违约损失率越低的信用本质属性的前提下,保证信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同等级的信用等级最优划分方法。
本发明所述方案实施步骤如下:
以中国某地区性商业银行分布在京津沪渝等 28 个城市的 1814
笔工业小企业贷款数据为实证样本,对本发明所述方案进行实证分析,具体步骤如下:
构建信用风险评级体系,包括建立信用风险评价指标体系、信用风险评价指标的赋权、建立客户信用风险评价方程以及划分客户信用等级;依次通过建立信用风险评价指标体系、信用风险评价指标的赋权和建立客户信用风险评价方程确定第
i 个客户的信用风险评价得分 S
i ,为信用等级的划分提供数据基础,对客户的信用风险评价得分
S
i 划分为 9 个信用等级;其中, i =1, 2, …n , n
为待划分信用等级的全部客户数目。
步骤 1 :确定信用风险评价得分 S
i
1) 建立信用风险评价指标体系:首先通过 Fisher
判别的方法在海选指标中遴选显著区分违约与非违约客户的指标;再通过相关分析法从上述显著区分违约与非违约客户的指标中删除反映信息重复的指标,得到信用风险评价指标体系;
2) 赋权信用风险评价指标:通过均方差方法对步骤 (1)
得到信用风险评价指标体系中指标赋权,指标的均方差越大,权重越大;
3)
建立客户信用风险评价方程:对信用风险评价指标体系中指标与指标的权重进行线性加权,建立客户信用风险评价方程
S
i =
∑ω
j
x
ij , 确定第 i
个客户的信用风险评价得分 S
i ;其中, ω
j -
第 j 个指标的权重, x
ij - 第 j 个指标下第 i
个客户的数值, i =1, 2, …n , j =1, 2, …m , n
为待划分信用等级的全部客户数目, m 为信用风险评价指标体系中指标的个数。
信用风险评价指标体系如表 1 第 2 列所示,指标权重如表 1 第 3 列 所示。
表 1 信用风险评价指标体系及指标权重
(1) 序号 | (2) 指标 x j | (3) 权重 ω j |
1 | X1 流动负债经营活动净现金流比率 | 0.035 |
2 | X2 主营业务收入现金比率 | 0.027 |
3 | X3 股东权益比率 | 0.031 |
… | … | … |
24 | X24 法律纠纷情况 | 0.175 |
25 | X25 抵质押担保得分 | 0.038 |
构建信用风险评价指标体系及确定指标权重是计算信用风险评价得分
S
i 的基础,并且具有若干种确定方法。
步骤 2 :数据导入
将 1814 笔贷款客户的信用风险评价得分 S
i
、应收未收本息 L
ki 、应收本息 R
ki 数据导入到
Excel 文件中,并按照信用风险评价得分 S
i 从高到低降序排列。
步骤 3 :信用等级划分
利用基于信用相似度最大的信用等级划分优化算法,对客户进行信用等级划分,并将划分结果直接显示于 Excel
界面。其中,信用评级结果包括:各信用等级的客户数 m
k 、各信用等级的违约损失率
LGD
k (k=1, 2, 3, 4, 5, 6, 7, 8, 9)
、各信用等级的信用风险评价得分区间、目标函数值,各信用等级违约损失率金字塔分布,如图 1 所示。
基于信用相似度最大的信用等级划分优化算法包括:
(1) 目标函数 1 :以同一等级中客户信用风险评价得分的组内离差最小,即: min f
1 =g
1(S
k ,S
ki ) ,其中,
S
k 表示第 k 个信用等级所有客户信用风险评价得分的均值,
S
ki 表示第 k 个信用等级中第 i 个客户的信用风险评价得分, k
=1, 2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, … 。
目标函数 1
确保信用风险评价得分越相近的客户越能划分为同一信用等级,避免将信用风险评价得分差异大的客户划分为同一信用等级引起信用风险评价得分区间长度过大,导致信用等级划分区间不具有区分度。
目标函数 2 :以不同信用等级中客户信用风险评价得分的组间离差最大,即: max f
2=g 2(S
k ,S)
,其中, S
k 表示第 k 个信用等级所有客户信用风险评价得分的均值, S
表示全部 9 个等级所有客户信用风险评价得分的均值, k =1, 2, 3, 4, 5, 6, 7, 8, 9 。
目标函数 2
确保不同信用等级间的信用风险评价得分差异尽可能大,避免信用风险评价得分区间长度过小而导致的客户的信用状况稍微发生变化,信用等级也将发生变化的过于敏感、稳定性不足的弊端。
(2) 约束条件 1 :信用等级从高到低违约损失率严格递增,即 0<LGD
1<LGD 2<
LGD 3 <LGD
4<LGD 5<LGD
6<LGD 7<LGD
8<LGD 9 ≤ 1 。
约束条件 1 通过设定信用等级从高到低违约损失率严格递增的约束,保证了信用评级结果满足 '
信用等级越高、违约损失率越低 ' 的信用本质属性,改变了现有评级体系可能导致的信用等级很高、但违约损失率反而不低的怪现象。
约束条件 2 :第 k 个信用等级违约损失率
LGD
k 计算的等式约束,即 LGD
k
=h( L
ki , R
ki )
;其中, L
ki 表示第 k 个等级第 i 个贷款客户的应收未收本息,
R
ki 表示第 k 个等级第 i 个贷款客户的应收本息, k =1,
2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, … 。
约束条件 2 通过贷款客户 的应收未收本息
L
ki 和应收本息 R
ik 的对比关系
,计算各信用等级贷款客户的违约损失率。保证了违约损失率的测算能真实的反映银行的损失。
应该指出 :如果不采用本发明中以信用风险评价得分为基础的目标函数,仅采用已有专利 (
专利号为: 201210201461.6 和 2012102001114.3)
的划分方法,会导致信用等级划分结果不具有评分区间稳定性的优点,即会导致信用风险评价得分区间长度长短不一、过大或过小。如果信用风险评价得分区间长度过小,客户的信用风险评价得分稍微变化,客户的信用等级也会发生变化,即信用风险评价得分区间过于敏感。如果信用风险评价得分区间长度过大,客户的信用风险评价得分发生较大变化,客户的信用等级也不会发生变化,即信用风险评价得分区间不具有区分度。
以中国某地区性商业银行分布在京津沪渝等 28 个城市的 1814
笔工业小企业贷款数据为实证样本,利用本发明所述等级划分方法进行信用等级划分,得到的基于信用相似度最大的信用等级划分结果,如表 2 所示。
表 2 各信用等级的评分区间及违约损失率
(1) 序号 |
(2) 信用等级 | (3) 评分区间 | (4) 评分区间长度 | (5) 违约损失率 LGD k |
1 | AAA | 73.41≤S≤100 | 26.59 | 0.130% |
2 | AA | 66.03≤S <73.41 | 7.38 | 0.263% |
3 | A | 60.11≤S <66.03 | 5.92 | 0.684% |
4 | BBB | 34.02≤S <60.11 | 26.09 | 6.040% |
5 | BB | 29.20≤S <34.02 | 4.82 | 9.543% |
6 | B | 27.28≤S <29.20 | 1.92 | 24.452% |
7 | CCC | 26.23≤S <27.28 | 1.05 | 33.868% |
8 | CC | 17.66≤S <26.23 | 8.57 | 35.448% |
9 | C | 0≤S <17.66 | 17.66 | 90.044% |
其中,表 2 第 3 列是每个信用等级的信用风险评价得分区间,表 2 第 4 列是根据第 3
列确定的信用风险评价得分区间长度。其中评分区间长度最小值是 1.05 ,是这 1814 个贷款客户中相邻两个客户信用风险评价得分差的均值 0.04 的 26
倍,因此信用风险评价得分区间具有一定的区分度。
以表 2 第 5 列的违约损失率 LGD
k 为横轴,表
2 第 2 列的信用等级 k 为纵轴,得到各信用等级违约损失率金字塔分布图,如图 1 所示。其中,图 1
满足信用等级越高、违约损失率越低的违约金字塔标准,这从表 2 第 5 列中不同等级的违约损失率数值可以看出。由表 2 第 4
列的信用等级评分区间长度、即相邻两个信用等级得分最大值之差的均匀分布结果可以看出,信用等级划分结果满足信用风险评价得分越相近的客户越易划分为相同等级、信用风险评价得分差异越大的客户越易划分为不同等级,即不同信用等级的信用风险评价得分区间长度分布稳定。
本发明尚有多种具体的实施方式,凡采用本发明所述 ' 基于信用相似度最大的信用等级最优划分方法 '
等同替换、或者等效变换而形成的所有技术方案,均落在本发明要求保护的范围内。
Claims (1)
1. 一种基于信用相似度最大的信用等级最优划分方法,其特征在于,步骤如下 :
步骤 1 :确定信用风险评价得分 S
i
步骤 2 :数据导入
将步骤 1 中得到待划分的所有客户的信用风险评价得分 S
i
、应收未收本息 L
ki 、应收本息 R
ki 的源数据导入到
Excel 文件中,按照信用风险评价得分从高到低降序排列;
步骤 3 :信用等级划分
利用基于信用相似度最大的信用等级划分优化算法,对客户进行信用等级划分,并将划分结果直接显示于 Excel
界面;
基于信用相似度最大的信用等级划分优化算法包括:
(1) 目标函数 1 :以同一等级中客户的信用风险评价得分组内离差最小,即: min f
1 =g
1(S
k ,S
ki ) ,其中,
S
k 表示第 k 个信用等级中所有客户信用风险评价得分的均值,
S
ki 表示第 k 个信用等级中第 i 个客户的信用风险评价得分, k
=1, 2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, … ;
目标函数 2 :以不同信用等级中客户信用风险评价得分的组间离差最大,即: max f
2=g 2(S
k ,S)
,其中, S
k 表示第 k 个信用等级中所有客户信用风险评价得分的均值, S
表示 9 个信用等级中所有客户信用风险评价得分的均值, k =1, 2, 3, 4, 5, 6, 7, 8, 9 ;
(2) 约束条件 1 :各信用等级从高到低违约损失率严格递增
即 0<LGD 1<LGD
2<LGD 3<LGD
4<LGD 5<LGD
6<LGD 7<LGD
8<LGD 9 ≤ 1 ;
约束条件 2 :第 k 个信用等级违约损失率 LGD
k
计算的等式约束,即 LGD
k =h( L
ki
, R
ki ) ;其中, L
ki 表示第 k
个 信用 等级中第 i 个客户的应收未收本息, R
ki 表示第 k 个 信用
等级中第 i 个客户的应收本息, k =1, 2, 3, 4, 5, 6, 7, 8, 9 , i =1, 2, …
;
通过确定步骤 3 所述目标函数 1 、目标函数 2 、约束条件 1 和约束条件 2
组成的多目标规划模型,通过多目标规划模型的求解得到最优的信用等级划分,使得信用等级划分结果在满足信用等级越高、违约损失率越低的违约金字塔标准下,保证信用状况相似的客户划分为同一个等级,信用状况差异大的客户划分为不同的等级。
2. 根据权利要求 1 所述的信用等级最优划分方法,其特征在于,
客户信用风险评价得分 S
i 的计算方式:
1) 建立信用风险评价指标体系:首先通过 Fisher
判别的方法在海选指标中遴选显著区分违约与非违约客户的指标;再通过相关分析法从上述显著区分违约与非违约客户的指标中删除反映信息重复的指标,得到信用风险评价指标体系;
2) 赋权信用风险评价指标:通过均方差方法对步骤 1)
得到信用风险评价指标体系中指标赋权,指标的均方差越大,权重越大;
3) 建立客户信用风险评价方程:对信用风险评价指标体系中指标与指标的权重进行线性加权,建立客户信用风险评价方程
S
i =
∑ω
j
x
ij , 确定第 i
个客户的信用风险评价得分 S
i ;其中, ω
j 第
j 个指标的权重, x
ij 第 j 个指标下第 i 个客户的数值,
i =1, 2, …n , j =1, 2, …m , n
为待划分信用等级的全部客户数目, m 为信用风险评价指标体系中指标的个数。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2016/079681 WO2017181346A1 (zh) | 2016-04-19 | 2016-04-19 | 基于信用相似度最大的信用等级最优划分方法 |
US15/765,584 US20180308158A1 (en) | 2016-04-19 | 2016-04-19 | An optimal credit rating division method based on maximizing credit similarity |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2016/079681 WO2017181346A1 (zh) | 2016-04-19 | 2016-04-19 | 基于信用相似度最大的信用等级最优划分方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2017181346A1 true WO2017181346A1 (zh) | 2017-10-26 |
Family
ID=60115549
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2016/079681 WO2017181346A1 (zh) | 2016-04-19 | 2016-04-19 | 基于信用相似度最大的信用等级最优划分方法 |
Country Status (2)
Country | Link |
---|---|
US (1) | US20180308158A1 (zh) |
WO (1) | WO2017181346A1 (zh) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110222129A (zh) * | 2019-06-17 | 2019-09-10 | 山东浪潮商用系统有限公司 | 一种基于关系型数据库的信用评价算法 |
CN111428982A (zh) * | 2020-03-18 | 2020-07-17 | 国网浙江杭州市临安区供电有限公司 | 一种基于大数据的优质光伏客户评价和筛选方法 |
CN112990685A (zh) * | 2021-03-10 | 2021-06-18 | 海南电网有限责任公司信息通信分公司 | 一种基于精准客户分群的差异化供电服务方法 |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110751403A (zh) * | 2019-10-21 | 2020-02-04 | 中国民航信息网络股份有限公司 | 信用评分方法及装置 |
CN111325620A (zh) * | 2020-02-17 | 2020-06-23 | 北京明略软件系统有限公司 | 信用评价模型的生成方法、装置及信用评价方法、装置 |
CN113450202A (zh) * | 2020-03-27 | 2021-09-28 | 台湾联合金融科技股份有限公司 | 信用评等系统及信用评等方法 |
CN112116402A (zh) * | 2020-09-28 | 2020-12-22 | 中国建设银行股份有限公司 | 信用风险客户级违约率确定方法及装置 |
CN112418987B (zh) * | 2020-11-20 | 2022-04-29 | 厦门大学 | 交通运输单位信用评级方法、系统、电子设备及存储介质 |
CN112465393B (zh) * | 2020-12-09 | 2022-07-08 | 南威软件股份有限公司 | 基于关联分析FP-Tree算法的企业风险预警方法 |
CN112686553A (zh) * | 2020-12-31 | 2021-04-20 | 民生科技有限责任公司 | 企业自动评级管理系统、方法、服务器及可读存储装置 |
CN112907035B (zh) * | 2021-01-27 | 2022-08-05 | 厦门卫星定位应用股份有限公司 | 基于K-means的交通运输主体信用评级方法及装置 |
CN112801542A (zh) * | 2021-03-09 | 2021-05-14 | 海南电网有限责任公司信息通信分公司 | 用电客户信用评估方法 |
CN113313574A (zh) * | 2021-06-09 | 2021-08-27 | 中关村科技租赁股份有限公司 | 一种科技租赁信用风险计量方法及系统 |
CN113421027A (zh) * | 2021-07-21 | 2021-09-21 | 北京优奥创思科技发展有限公司 | 一种基于数据运算对客户消费行为评级的方法 |
CN115689394A (zh) * | 2022-12-29 | 2023-02-03 | 上海意静信息科技有限公司 | 一种社会单位消防信用等级评价方法及评价系统 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20010039523A1 (en) * | 2000-04-28 | 2001-11-08 | The Tokio Marine And Fire Insurance Co., Ltd | System and method for supporting provision of rating related service |
CN102163205A (zh) * | 2010-02-21 | 2011-08-24 | 施章祖 | 一种类似客户群体的自动挖掘系统 |
CN102629296A (zh) * | 2012-02-29 | 2012-08-08 | 浙江工商大学 | 一种基于灰色模糊的企业信用评价方法 |
CN102779317A (zh) * | 2012-06-18 | 2012-11-14 | 大连理工大学 | 基于信用等级与违约损失率匹配的信用评级系统与方法 |
CN102800016A (zh) * | 2012-06-18 | 2012-11-28 | 大连理工大学 | 基于信用等级与违约损失率匹配的信用评级调整方法 |
CN105956915A (zh) * | 2016-04-19 | 2016-09-21 | 大连理工大学 | 基于信用相似度最大的信用等级最优划分方法 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070288360A1 (en) * | 2006-05-03 | 2007-12-13 | Joseph Guy Seeklus | Systems and methods for determining whether candidates are qualified for desired situations based on credit scores |
-
2016
- 2016-04-19 US US15/765,584 patent/US20180308158A1/en not_active Abandoned
- 2016-04-19 WO PCT/CN2016/079681 patent/WO2017181346A1/zh active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20010039523A1 (en) * | 2000-04-28 | 2001-11-08 | The Tokio Marine And Fire Insurance Co., Ltd | System and method for supporting provision of rating related service |
CN102163205A (zh) * | 2010-02-21 | 2011-08-24 | 施章祖 | 一种类似客户群体的自动挖掘系统 |
CN102629296A (zh) * | 2012-02-29 | 2012-08-08 | 浙江工商大学 | 一种基于灰色模糊的企业信用评价方法 |
CN102779317A (zh) * | 2012-06-18 | 2012-11-14 | 大连理工大学 | 基于信用等级与违约损失率匹配的信用评级系统与方法 |
CN102800016A (zh) * | 2012-06-18 | 2012-11-28 | 大连理工大学 | 基于信用等级与违约损失率匹配的信用评级调整方法 |
CN105956915A (zh) * | 2016-04-19 | 2016-09-21 | 大连理工大学 | 基于信用相似度最大的信用等级最优划分方法 |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110222129A (zh) * | 2019-06-17 | 2019-09-10 | 山东浪潮商用系统有限公司 | 一种基于关系型数据库的信用评价算法 |
CN110222129B (zh) * | 2019-06-17 | 2023-09-22 | 浪潮软件科技有限公司 | 一种基于关系型数据库的信用评价算法 |
CN111428982A (zh) * | 2020-03-18 | 2020-07-17 | 国网浙江杭州市临安区供电有限公司 | 一种基于大数据的优质光伏客户评价和筛选方法 |
CN112990685A (zh) * | 2021-03-10 | 2021-06-18 | 海南电网有限责任公司信息通信分公司 | 一种基于精准客户分群的差异化供电服务方法 |
Also Published As
Publication number | Publication date |
---|---|
US20180308158A1 (en) | 2018-10-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2017181346A1 (zh) | 基于信用相似度最大的信用等级最优划分方法 | |
CN108846547A (zh) | 一种动态调整的企业信用风险评估方法 | |
CN102779317B (zh) | 基于信用等级与违约损失率匹配的信用评级系统与方法 | |
WO2019140675A1 (zh) | 基于逼近理想点违约鉴别能力最大的信用评级最优权重向量的方法 | |
JP2001117973A (ja) | 不動産評価装置、不動産評価方法、不動産評価プログラムを格納する記録媒体および不動産情報提供システム | |
CN114202103A (zh) | 一种基于机器学习的节假日景区客流预测方法 | |
Turrell et al. | Using job vacancies to understand the effects of labour market mismatch on UK output and productivity | |
CN108876487A (zh) | 一种基于大数据及智能决策机制的工业地块估计方法 | |
CN109102377A (zh) | 一种基于区块链技术的不良资产拍卖处置平台 | |
Liu et al. | A Bayesian approach to residential property valuation based on built environment and house characteristics | |
CN107578165A (zh) | 基于粗糙集中简约算法的银行营销管理方法及系统 | |
CN112801779A (zh) | 一种企业融资信用评价系统及方法 | |
CN105956915A (zh) | 基于信用相似度最大的信用等级最优划分方法 | |
Brooks | Why is unemployment high in the Philippines? | |
CN113988930A (zh) | 商业房地产人工智能估价系统 | |
Kathiravan et al. | Does temperature influence the carbon index? Evidences from India | |
Kim et al. | Working from home, commuting time, and intracity house‐price gradients | |
CN109886557A (zh) | 一种基于大数据和网格化的订单分发方法及系统 | |
CN114880923A (zh) | 一种融合电力数据的企业信用预测方法 | |
Qu | A decision tree applied to the grass-roots staffs' turnover problem—take CR Group as an example | |
CN111858639B (zh) | 用于风控管理的外部数据管理系统及方法 | |
TRAN et al. | The effects of corporate governance on segment reporting disclosure: A case study in Vietnam | |
Vetoshkina et al. | Financial Statements Fraud Detection: Analysis of Beneish and Roxas Models Applicability (Russian Approach) | |
Zhang et al. | Credit Evaluation of the Highway Construction Enterprises Based on the Combination Weighting-Grey Correlation TOPSIS Model | |
Aliyu et al. | Influence of aircraft noise on residential property values: Evidence from current literature |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 15765584 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 16898936 Country of ref document: EP Kind code of ref document: A1 |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 16898936 Country of ref document: EP Kind code of ref document: A1 |