大数据环境下的企业违约预警研究：基于参数优化的视角

企业违约预警是通过对违约风险的判别，最大限度地减低企业违约所造成的损失。违约预警体系的合理与否，对商业银行贷款发放、债券投资、股票投资和供应链金融的授信等都极为重要。本项目采用结构化和非结构化的海量数据，通过支持向量机、决策树等大数据挖掘方法，寻找能使违约预警效果最优的一组参数，建立违约预警模型。内容包括：(1)最优指标组合的遴选。通过企业违约状态与指标组合数据的函数关系，遴选违约鉴别能力最大的指标组合。(2)最优权重向量的确定。通过构建评价结果与权重向量之间的数学规划模型，反推一组违约预警能力最强的指标权重向量。(3)最优违约判别临界点的选取。基于违约预测方程的预测结果与不同违约判别临界点之间的函数关系，以预测方程精确度最大或错判损失最小为目标，反推最优的违约预测临界点。本研究在上述参数最优化的基础上，建立企业违约预警模型。

Enterprise default early warning is the identification of the default risk to minimize the losses caused by the enterprise's default. The rationality of the early warning system is extremely important to commercial bank loans, bond investment, stock investment and credit extension of supply chain finance. Our project takes mass data including structured data and unstructured ones to construct the default prediction models by selecting the optimal parameters with greatest discriminant performance based on big data mining methods. It includes three folds. First, select the optimal indicator combination. Our project selects the indicator combination with the largest default discrimination and prediction capability by building the functional relationship between the default status and the indicator combination data. Second, determine the optimal weight vector. Our project reversely infers the weight vector with the greatest default prediction ability by building the functional relationship between the evaluation results obtained by the default prediction model and the weight vector. Third, select the optimal default prediction threshold. Our project selects the default discriminant threshold with maximum accuracy or the minimum loss by building the functional relationship between the evaluation results obtained by the default discriminant equation and the corresponding thresholds. Our study will use the optimal paraments above to construct the default early warning model for enterprises.