基于不均衡支持向量机的小企业信用风险评价理论与模型

Small business credit risk evaluation is an issue of banks' risk management, and is also related to the economic and social stability..For small business loans, the default samples are far less than the non-default samples. Also, the impact on banks of default customers' misjudgment is much larger than the impact on banks of non-default customers' misjudgment. In this case, this research will identify small business credit risk element, use the unbalanced support vector machine to evaluate the credit risk, analysis the impact of external shocks to small business credit risk..According to the impact of specific indicator to small businesses customer defaults or not, the invalid indicators which can not divide the default customers and the non-default customers will be removed. Then the factors of small business credit risk will be identified. .According to the principle that the greater the gap between the default customers and the non-default customers, the greater the weight, we will establish an evaluation equation with significant distinction based on unbalanced support vector machine.Using the discriminated ability as the evaluation model test standard, we will change the phenomenon that the overall accuracy is high, but the default samples' accuracy is not high, which is cased by the unbalanced data. Giving greater weight to the default sample than the non-default sample, we will reflect the principle of the default customers have a greater impact on bank..By establishing the function of external shocks such as the macro environment, industry environment and the small business credit risk evaluation score, we will reveals the effect of external shocks in past and present to the credit risk of small business in the future..This research aims at establishing a group of small business credit risk evaluation methods based on unbalanced support vector machine and extending a new vision on small business credit risk evaluation theory and models.

小企业信用风险评价既是银行风险管理问题，又事关经济社会稳定。针对小企业贷款中，违约样本远少于非违约样本、违约客户误判对银行影响较大的现实，鉴别小企业信用风险的关键影响要素，采用不均衡支持向量机进行评价，分析外部冲击对小企业信用风险的影响程度。根据特定指标对小企业违约状态的影响，保留能清晰鉴别客户违约与否的指标。根据样本差距越大、权重越大的赋权思路，采用不均衡支持向量机构建具有显著区分违约能力的评价方程。以违约判别能力作为模型检验标准，改变样本数据不均衡所导致的样本总体精度很高、违约样本精度反而不高的现象。给违约样本赋予大于非违约样本的权重，体现违约样本误判对银行影响较大的原则。通过宏观环境、行业环境等外部冲击与小企业信用风险评价得分的函数关系，揭示过去和现在的外部冲击对未来小企业信用风险的影响。本研究致力于提供一类基于不均衡数据的小企业信用风险评价方法，补充改善小企业风险评价理论和模型。

小企业占我国企业总数99%以上，提供75%以上的城镇就业岗位。由于其资金实力弱，财务制度不完善，小企业融资难、贷款难的问题相当严重。小企业信用风险评价不仅是一个银行风险管理问题，而且事关经济社会稳定。针对小企业贷款样本中，违约样本远少于非违约样本、且违约客户误判代价较大的现实，对小企业信用风险评价体系进行了研究。.首先，以我国某地方性大型商业银行可收集的小企业信贷数据为基础，结合国内外机构的高频指标，通过偏相关分析，剔除偏相关系数大的、反映信息重复的指标；通过有无评价指标分析，遴选对贷款小企业违约状态有影响的指标。在有无指标分析中，通过违约样本误判率、非违约样本和违约样本的平均识别率、样本总体正确率等一系列违约状态识别比率，提高了小样本违约指标对评价方程的影响程度，解决了样本不均衡下指标重要程度识别的问题。最终构建了包括相关行业从业年限、主营业务收入现金比率、抵质押等28个指标的小企业信用风险评价指标体系。并构建了工业类、服务业类小企业的信用风险评价指标体系。.其次，采用不均衡支持向量机对小企业信用风险评价指标进行赋权，进而构建了能有效区分违约客户、非违约客户的评价模型。根据有无特定评价指标、特定评价指标数值变化大小对贷款小企业违约状态的影响程赋权；反映了对违约状态影响越大、评价指标权重越大的赋权思路。以违约样本正确识别率、违约样本的准确率与查全率等因素作为基于支持向量机赋权模型的度量标准，改变样本数据不均衡所导致的样本总体精度很高、违约样本精度反而不高的现象。结果表明：行业景气指数、资本固定化比率、净利润现金含量等评价指标对小企业信用风险的评价影响较大。.再次，采用Logistic回归模型分析、多元线性回归分析小企业负责人的背景特征、宏观经济环境对小企业违约行为的影响。结果表明：小企业负责人的居住状况、在本地居住年限、汽车及不动产价值总额与小企业信贷违约风险之间具有显著负相关关系。GDP增长率越高，贷款小企业越不可能发生违约行为、且违约损失率也越小。此外，GDP增长率对违约影响因素与企业违约风险之间的关系有调节作用，即：GDP增长率越高，流动资产周转速度、总资产增长率对企业是否违约和违约损失率的影响作用越大。.综上，本项目围绕小企业信用风险评价及影响机理逐步展开，提供了一类基于不均衡数据的小企业信用风险评价方法，补充完善了小企业信用风险评价理论与模型。