【作者】 姜明辉；

【导师】 王雅林；

【作者基本信息】 哈尔滨工业大学 ， 技术经济及管理， 2006， 博士

【摘要】 随着中国经济的快速发展,信用消费增长迅速,住房按揭、汽车贷款、教育贷款、信用卡等各种个人消费贷款的规模在迅速扩大。在消费信贷热不断升温的形势下,各商业银行均把发展消费贷款作为未来发展战略的重要组成部分。但是目前国内商业银行对消费贷款的风险管理水平较低,管理手段与方法均较落后,在消费信贷的发放过程中,仍然采用传统的信用分析方法来评价消费信贷申请者的信用状况及还款能力,在个人信用评估方法上仍然没有形成稳健可靠的模型。而目前学术界虽然提出了包括数理统计、人工智能等多种信用评估模型,但是这些方法都无法同时达到精确性和稳健性的最优状态。本文在分析国内外个人信用评估发展历史及其方法应用现状的基础上,指出目前个人信用评估方法指标体系混乱、非系统性、精确度与稳健性无法兼顾等问题。针对指标体系混乱问题,利用货币效用曲线分析影响个人消费贷款履约行为的关键性因素,参考国内外已有的指标体系,构建适合中国国情的商业银行个人信用评估指标体系。针对非系统性问题,结合目前我国商业银行业务系统中所获取的实际数据情况,确定评估中所使用的指标,并就指标赋值、标准化、数据缺失、违约标准确定等数据处理过程提出了相应方法并加以应用,在目前常用的抽样方法中选择合理的方法加以应用,确定样本容量并进行分组,从而保证数据样本的可靠性,从而建立系统性的评估思路。针对精确度与稳健性无法兼顾问题,利用组合预测的思想,综合考虑权重非负约束、目标函数的最优解准则、可变权重、计算的复杂性、预测精确度等因素,建立了个人信用评估的线性组合预测模型、变权组合预测模型和非线性组合预测模型,提出了应用各模型进行预测的求解方法。分别就个人信用评估的统计模型和神经网络模型进行筛选、应用和比较,选择有效的单一模型对基于统计方法的组合预测模型、基于统计方法和神经网络的组合预测模型、基于神经网络的组合预测模型进行应用,对应用的结果进行比较分析。通过研究发现,统计模型在分类精确度方面不如神经网络模型,但神经网络模型在稳健型性方面较统计模型差。组合预测模型比单一模型在分类精确度以及模型稳健性方面均具有优势。而组合模型之间比较的结果,综合考更多还原

【Abstract】 With China’s rapid economy development and rapid growth of consumer credit, mortgage and car loans, education loans, credit cards and other consumer loans are increasing rapidly. In condition of continuously heated consumer credit, commercial banks take consumer loans as an important component of their future development strategy. However currently domestic commercial banks’ level of risk management on consumer loans is low and their management methods were backward. In consumer credit payment process, they still use traditional methods of credit analysis to evaluate the consumer credit applicants’ credit standing and their ability to repay. There has been no reliable model in personal credit scoring methods. Although in current academic community, kinds of credit scoring model are presented including mathematical statistics, artificial intelligence, these are impossible to achieve optimum state of both accuracy and stability.Based on the analysis about the development history of personal credit scoring history and the application status of personal credit scoring models, this dissertation points out the status of personal credit scoring in China, such as the system chaos and inefficiency as well as non-compatibility of accuracy and stability in personal credit index system. Concerning on current chaos personal credit index, a personal credit index of commercial banks has been established which references existed index systems in and abroad and applies money effect curve to analyze crucial factors effecting personal consuming loan default behavior. Focusing on non-systemic problems, we combine real date in current commercial banks, establish evaluation index, bring forward and applying methods corresponding to data processing such as index evaluation, initialization, data absence and identification of default standards. Choosing sensible methods in current common-used sampling methods, identifying sample capacity and then grouping samples we ensure the credibility of data samples and establish systemic evaluation methods as well. As for the non-capability of accuracy and stability, we employ the combination theory and make a comprehensive consideration on non-negative weights, optimized solution of objective functions,更多还原