基于蒙特卡洛模拟的个人信用评分样本集优化研究

发布时间：2018-04-23 09:39

本文选题：信用评分 + 样本偏差　；参考：《哈尔滨工业大学》2014年硕士论文

【摘要】：随着商业银行的快速发展，信贷业务在其发展进程中的地位日益显著，信贷风险水平的控制直接影响到商业银行整体的进展。因此，对个人信贷风险的识别和评估的准确与否，就成为了商业银行能否合理控制风险的至关因素。过去的研究主要集中在模型精度的提高对评分方法进行挖掘及优化，而忽略了“拒绝推论”即样本偏差这一重要问题。由于很多评分机构只能将接受样本的部分数据通过信用模型来预测贷款申请者违约与否，导致出现样本偏差问题进而影响信用评分模型的有效性。因此，在信用评估范畴内，样本偏差问题是一个急需处理的问题。本文首先对样本偏差及其纠正技术进行了分析。提出了利用蒙特卡洛生成随机数的方法产生样本，加入到原始样本中形成新样本集从而优化样本集，利用加入生成样本后的优化样本集来解决样本偏差这一问题。在产生样本方面，本文对生成样本的生成方法进行了阐述及选择，并对个人信用评分指标体系进行构建，确定商业银行个人信用评分指标，通过对商业银行原始样本的指标进行分析，确定其分布规律及内在联系，产生指标数据，，同时对数据相应的还款情况进行分类，运用软件进行训练形成生成样本。并对生成样本加入原始样本中好坏客户比例及生成样本和原始样本的搭配比例进行了构造及说明，最终形成优化样本集。最后，利用优化样本集对评分领域单一模型和组合预测模型进行了效果检验。检验结果表明，将优化样本集作为个人信用评分模型的样本数据不仅能使其预判精度明显提高，同时也能使得样本偏差问题得到很好的解决。
[Abstract]:With the rapid development of commercial banks, credit business is playing an increasingly important role in the development process. The control of credit risk level directly affects the overall progress of commercial banks. Therefore, the identification and evaluation of personal credit risk is the most important factor for commercial banks to reasonably control the risk. In the past, the research focused on improving the accuracy of the model to mine and optimize the scoring method, while ignoring the important problem of "reject inference", that is, sample deviation. Because many rating organizations can only use credit model to predict loan applicant default or not, the sample deviation problem will affect the validity of credit rating model. Therefore, sample deviation is an urgent problem in credit evaluation. In this paper, the sample deviation and its correction technique are analyzed. This paper presents a method of generating samples by using Monte Carlo to generate random numbers, which is added to the original samples to form a new sample set to optimize the sample set. The problem of sample deviation is solved by using the optimized sample set after adding the generated sample. In the aspect of producing samples, this paper expounds and selects the method of generating samples, and constructs the index system of personal credit score, and determines the index of personal credit score of commercial banks. Through the analysis of the index of the original sample of commercial bank, the distribution law and internal relation are determined, and the index data is produced. At the same time, the corresponding repayment situation of the data is classified, and the training software is used to form the generated sample. Finally, the optimal sample set is formed by constructing and illustrating the proportion of good or bad customers added to the original sample and the collocation ratio between the generated sample and the original sample. Finally, the single model and combined prediction model in scoring field are tested by the optimized sample set. The test results show that using the optimized sample set as the sample data of personal credit scoring model can not only improve the precision of prediction, but also solve the problem of sample deviation.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F224;F830.5

【参考文献】