【作者】 史宁；

【导师】 姜明辉；

【作者基本信息】 哈尔滨工业大学 ， 国际贸易学， 2008， 硕士

【摘要】 随着我国消费信贷市场的不断发展,个人信用得到了空前的重视。我国目前尚未建立起完善的个人征信体系,商业银行目前没有一套科学合理的个人信用评估指标和方法。在这种情况下,加强对个人信用评估方法的研究,对于商业银行降低消费信贷风险,减少不良贷款比率,扩大消费信贷的积极性,促进国民经济快速稳定发展具有积极意义。本文在对国内外相关研究成果进行分析的基础上,首先阐述了个人信用评估中用到的三种统计方法的基本原理并分别建立了基于这三种统计方法的个人信用评估模型并进行了检验,随后对组合预测原理及其权重求解的方法进行了阐述,最后构建了基于线性回归、Logistic回归和Probit回归三种统计方法的非负约束权重的组合预测模型,并将组合预测模型应用到个人信用评估中。在非负约束权重的组合预测模型的权重求解计算中,引入了较传统解法更有效、结果更合理的二次规划的神经网络解法,并且利用遗传算法对非负权重进行了求解。在最后的模型应用中,将非负约束权重的组合预测模型与单一模型进行对比,结果表明:从总分类准确率和总错分率来看,三种单一统计模型及组合预测模型的分类准确率都在90%以上,其中非负约束权重的组合预测模型的总分类效果最好,与此同时非负约束权重的组合预测模型的第一类错分率虽然不是最低,但在第二类错分率上具有优势,因此可以判定非负约束权重的组合预测模型的分类综合效果高于单一模型。将利用二次规划算法求解的非负约束权重的组合预测模型与利用遗传算法求解得到的非负约束权重的组合预测模型进行对比表明,二次规划算法中线性回归模型的权重为0,作为冗余方法被剔除,并且前者的分类效果优于利用遗传算法求解得到的结果。由此得出,利用二次规划方法求解得到的非负约束权重的组合预测模型在个人信用评估中是具有应用价值的结论。更多还原

【Abstract】 With the development of consumer credit in domestic commercial banks, personal credit becomes more and more important. At present there is no perfect personal credit system established in China, and there are no scientific and reasonable methods or indicators applied in personal credit assessment. In this situation, researches on personal credit scoring are critical for the commercial banks to reduce the risk of personal credit and the ratio of non-performing loan. It also has a positive significance for the commercial banks to expand the consumer credit market which can accelerate and stabilize the development of the national economy.Based on analysis of domestic and overseas in personal credit scoring, this dissertation firstly expatiates on the fundamental principles of the three statistical methods to be used in this domain and then establishes and tests the three models in personal credit scoring respectively. Then it elaborates the principles of combining forecasts as well as its methods of weights calculation. At last, after establishing a combined forecast model with non-negative combining weight constraint through multi-linear regression, logistic regression and Probit regression, this dissertation applies the combined model to personal credit scoring. In the weights calculation of the combination forecasting model, the dissertation uses the method Genetic Algorithms and introduces a new Neural Networks method of quadratic programming, which is more effective and reasonable. Through the contrast of the combining model with non-positive restriction on weights and the single ones in the final application, the result shows that the general accuracy of the models are all above 90%, while the effect of the combination forecasting model is better. At the same time, in the result of the combination forecasting model, the error rate of the first case is not the lowest, but its error rate of the second case has a comparative advantage. Therefore the combination forecasting model with non-negative combining weight constraint is generally better than the single ones. Through the contrast between the results of the two methods it shows that the weight of Multiple Linear Regression in quadratic programming is 0, so it is rejected. And the effect of classification based on the method of quadratic programming is better. Therefore it can be concluded that the combination forecasting model using Neural Networks of quadratic programming has an application value in personal credit scoring.更多还原