基于智能优化的信用风险评估方法与综合集成预警模型研究

With current status of China's bank credit, we shall investigate the basic theories of credit risk management and novel approaches of credit risk assessment and early warning， so as to provide theoretical support of constructing the credit risk warning system. On the basis of comprehensive analysis of the factors which may influence credit risk, we aim to customer credit risk including single and group customer risks in China's banking industry, build multi-dimension, multi-level early warning indicator system, and then propose customer risk warning model. After that we shift another important issue in China's bank industry, bank loan portfolio risk management and early warning, where risk optimization decision model will be intensively studied and the problem of intelligent computation based optimization solving will be in-depth investigated. These progresses promote us to design optimal integration strategy in a hybrid manner of qualitative and quantitative analysis so as to establish intelligent integrated risk warning model. The significance of this project is two-fold and summarized as follows. Theoretically, a novel methodology of intelligent computation based credit risk evaluation and early warning will enrich and improve credit risk management theory to accommodate to increasing trend of globalization. In application, intelligent integrated credit risk early warning system to be developed in the project is capable of providing near real-time warning of credit risk, which is currently extreme important to China's banking industry in effective control of credit risk.

本项目从我国银行信贷现状出发，深入研究信用风险管理基本理论和信用风险评估与预警的新方法，为信用风险预警系统的构建提供理论支撑。在全面分析信用风险影响因素的基础上，针对我国银行业在风险管理中的客户信用风险，包括单一客户风险和集团客户风险，构建多维度、多层次风险预警指标体系，提出客户风险预警模型；针对银行贷款组合风险，建立风险优化决策模型，研究基于智能计算的最优求解方法；在此研究基础上，设计最优集成策略，采用定性与定量相结合的方式，建立智能化综合集成风险预警模型。本项目研究的意义在于：1）理论上，提出一套新的信用风险评估与预警方法，丰富和完善信用风险管理理论；2）应用上，开发智能化综合集成信用风险预警系统，提供实时预警结果，为我国银行业有效控制信用风险提供决策支持。

本项目从我国主要银行的企业信贷数据出发，结合银行业务知识，利用金融风险管理基本理论和统计学方法，在全面分析客户风险影响因素的基础上，综合客户信贷数据、客户关联关系数据，宏观数据以及其他渠道的风险信息，构建了多维度、多层次的风险预警指标体系，筛选出重要的风险预警指标，并分析客户的风险特征，包括:区域风险特征、行业风险特征、跨行风险特征、关联风险特征和担保风险特征。在此基础上，采用统计方法、人工智能方法以及面板模型分别构建了单一客户及集团客户的风险预警模型以及区域和行业的风险预警模型。此外，在微观层面风险预警的基础上，采用复杂网络分析的方法对客户关联关系数据进行研究，构建了总网络、节点以及连边多层次的分析框架，并刻画了客户风险传染、区域行业的风险传播及其对系统性风险的影响。同时，采用事件分析方法，考虑不同信息触发的客户风险预警事件分析，验证和拓展原有的风险预警模型。在上述基础之上，构建了以识别与计量为基础，系统工程化以及全面风险管控的客户风险预警管理体系。该项成果是关于银行客户风险早期识别的重要探索、是现有信用风险管理体系的重要补充，对于推动客户风险的精细化管理管理具有重要意义。