违约损失率视角下小微企业信用评级研究

Credit rating of small and micro enterprises is an important means to alleviate the difficulty and cost of financing. Rationally evaluate the credit risk of small and micro enterprises can promote financial development and increase employment. In view of the fact that bank credit rating not only focus on default probability, but also on the key parameter of loss given default (LGD). Different from the traditional credit rating based on default state and neglects customer default loss, this project uses LGD as the standard to study the credit risk rating of small and micro enterprises. The convolutional neural network (CNN) model is applied to establish the correlation between the index data and LGD, and identify the key default feature that has an impact on LGD. A non-linear programming model is constructed to determine the default probability by giving the default customer a greater weight for the loss function than the non-default customer, and a credit evaluation model of small and micro enterprises under unbalanced samples is established. A non-linear programming model is established to determine the credit grade and corresponding LGD of micro-individuals, which minimize the intra-group deviation of default probability and LGD in each credit grade as the objective function, and the strict increment of LGD is taken as the constraint condition. The research improves the theory and method of credit risk management, and establish a credit rating model of small and micro enterprises based on LGD. It also helps investors and banks effectively identify the credit risk of small and micro enterprises.

合理评价小微企业的信用风险，有利于改善小微企业融资难、融资贵的现状，促进金融发展和增加就业。银行信用评级不仅注重违约概率、还要注重违约损失率这一现实需求。不同于传统以违约状态为标准、忽略客户违约损失的信用评级，本项目以违约损失率为标准研究小微企业信用风险评级。通过卷积神经网络模型建立指标数据与违约损失率之间的关联关系，识别对违约损失率影响显著的关键违约特征。通过对违约客户的损失函数赋予较大权重的思路构建非线性规划模型确定客户的违约概率，建立非平衡样本下小微企业信用评价模型。通过以每一信用等级中违约损失率和违约概率的组内离差最小为目标函数，以违约损失率严格递增为约束条件，建立非线性规划模型确定小微企业的信用等级。本研究完善了信用风险管理的评级理论与方法，建立了以违约损失率为标准的小微企业信用评级模型；有助于投资者和银行有效识别小微企业的信用风险。

基于项目申请书，本项目的研究主要包括五个部分：一是违约特征的识别。将同一指标划分为不同特征，通过对不同特征客户违约损失率均值差的显著性检验标准甄别关键违约特征，确定违约风险最大的贷款客户特征，抓住信用风险管理的关键，开拓信用风险评级理论的新思路。二是小企业信用风险评价指标体系的建立。提出了通过对数似然函数值衡量指标及指标体系违约判别能力这一新的标准，以对数似然函数值最大为目标函数，构建0-1整数规划，通过遗传算法求解出违约鉴别能力最大的小企业信用风险最优评价指标体系。三是针对数据非平衡问题的研究。提出了一种新的混合模型框架。该框架是在Logistic回归和反向传播神经网络与安全水平SMOTE相结合的基础上开发的。四是多维数据下小企业违约风险过程性评价研究。分别从增加新的准则和增加新的指标两个特征增加的角度，提出信贷过程中，违约风险评价构型构建的四种方式，并以中国的小企业实际贷款数据为对象，通过建立神经网络模型进行实证研究。五是信用等级划分。挖掘标准普尔和穆迪的信用等级规律，提出违约概率服从指数分布的信用等级最优划分方法。.理论成果方面，共发表和录用6篇学术论文。SSCI检索论文2篇，SCIE检索论文1篇，CSSCI检索论文2篇，国家自然科学基金委员会管理科学部认定的A类重要期刊上发表和录用论文3篇。完成学术专著1部。人才培养方面，项目主持人晋升副教授。依托本项目培养硕士生2名，本科生12名。.综上，本项目已达成预期研究目标。在违约损失率背景下对信用评级的研究取得了突破，拓展信用风险管理研究。