普惠金融视角下基于违约损失显著判别的农村个体工商户信用评价研究

There is a disadvantage of existing credit evaluation methods which are used to identify a loan default status. The methods fail to differentiate the levels of credit risk between customers who have “the same default status but with different levels of loss given default (LGD)”. To overcome this drawback, this study proposes a novel credit evaluation system (DISRC), through the steps of Data Driven (D), Indicator Selection (I), Credit Scoring (S), Credit Rating (R) and Default Characteristic Mining (C). The DISRC system can significantly distinguishthe credit ratings of rural individually owned enterprises with different default losses, reveal the relationship between evaluation indicators and default losses, and identify key determinants of customers’ credit risk. To build up the credit evaluation system, we firstly use fuzzy C-means to remove indicators reflecting redundant information then use fuzzy rough sets to filter indicators that have a significant impact on rural individually owned enterprises’ default losses. Secondly, we calculate the weights of the indicators based on the idea that the stronger the indicators' ability to identify the default losses of individually owned enterprises, the bigger the indicators' weights should be allocated. Thirdly, under the multiple constraints of inclusive finance, commercial sustainability and “loss - rating” matching, we establish a credit model which is suitable for small samples. Finally, we identify the key attributes that have the greatest impact on the default loss of rural individually owned enterprises from the same indicator’s various attributes. This will help rural individually owned enterprises to find out the key factors which influence their credit risk and improve their credit worthiness. This study offers a theoretical support for commercial banks to promote inclusive finance and provide financial support for the rural revitalization strategy, and explores new ideas for agricultural-related credit risk management.

针对现有以违约状态鉴别为目标的信用评价，无法区分“违约状态相同、但违约损失不同”客户信用风险差异的弊端，提出以违约损失显著鉴别为目标，通过数据驱动(D)、指标遴选(I)、得分求解(S)、等级划分(R)、违约特征挖掘(C)，构建显著区分农村个体工商户违约损失大小的DISRC信用评价体系，揭示评价指标与违约损失之间的规律性联系，甄别信用评价关键特征。(1)利用模糊C均值剔除反映信息冗余指标，利用模糊粗糙集筛选对商户违约损失判别影响显著的指标；(2)依据指标对商户违约损失判别能力越强、权重越大的思路，求解指标权重；(3)在普惠金融、商业可持续和“损失-等级”匹配约束下，建立小样本信用等级划分模型；(4)在同一指标不同“特征属性”中，遴选对商户违约损失影响最大的属性，找准影响商户信用风险的关键，提升其信用水平。为涉农商业银行推行普惠金融，践行金融支持乡村振兴提供依据，开拓涉农信用风险管理新思路。

涉农信用评价体系不完善是农村个体工商户等涉农信贷主体受到信贷配给的主要原因。为解决这一问题，本项目通过数据驱动（D）、指标遴选（I）、得分求解（S）、等级划分（R）、违约特征挖掘（C），构建了能显著区分农村个体工商户违约风险的DISRC信用评价体系，揭示了农村个体工商户信用评价指标特征与其违约风险之间的规律性联系，为完善涉农微观信贷主体信用风险评价理论，缓解农村个体工商户融资约束，为助力涉农金融机构实践推行普惠金融提供理论依据和经验证据。.主要完成工作：（1）通过项目合作和实地调研，建立了涵盖29个省10483个商户的信用风险评价数据库。（2）针对信用评价指标数据维度高、缺失严重、文本与数值信息混存等问题，利用人工智能鲸鱼优化算法，以Kolmogorov-Smirnov统计量最大为目标函数，建立了高维数据信用评价指标筛选模型，遴选出可显著鉴别农村个体工商户违约风险的评价指标。（3）针对信用评价中非违约样本多、违约样本少的数据不均衡特征，采用LDAMCE（标签分布交叉熵损失函数）修正BP神经网络（BPNN）中的交叉熵函数，构建了BPNN-LDAMCE信用评级模型，弥补了信用评价模型对非违约样本识别过度、违约样本识别度偏低的不足。（4）在普惠金融、商业可持续和“信用等级越高、违约损失越低”的“损失-等级”匹配约束下，建立了涉农农村个体工商户信用等级划分模型。（5）在同一指标不同“特征属性”中，遴选对客户信用水平影响最大的属性，找出了影响其信用风险的关键特征。.围绕上述研究，在《数量经济技术经济研究》《农业技术经济》《Annals of Operations Research》《International Journal of Forecasting》等期刊发表论文42篇，在《中国农业出版社》出版学术专著1部，联合主办The Impact of Fintech and Climate Change on Agricultural Finance and Investment Conference等国际会议3次；基于研究成果，积极建言咨政，呈送的研究报告得到中央农办秘书局局长吴宏耀同志，陕西省委常委、常务副省长王晓同志的肯定性批示，相关研究被新华社、信用中国官网等媒体报道转载，取得了良好的社会影响。

内容获取失败，请点击重试