基于多目标优化加权软投票集成算法的信用债违约预警研究

Research on credit bond default warning based on multi objective optimization weighted soft voting integration algorithm

为了提高信用债违约预测的准确性和稳定性,便于金融风险管理,以2014年1月1日—2021年12月31日的信用债为研究对象,提出一种基于多目标优化的加权软投票集成算法。该算法通过计算每个基分类器的模糊密度来量化其识别能力,并使用多目标粒子群算法来求解基分类器的权重。将所提算法与其他单一分类器如支持向量机、逻辑回归、高斯贝叶斯、MLP,以及其他集成算法如投票类集成算法(voting)和stacking算法进行比较,采用期望PFI算法进行特征重要度分析。结果表明,加权软投票集成算法在信用债违约预测中表现出色,不仅提升了单一算法的性能,且相对于其他集成算法,具有更高的准确性、精确度和AUC值。违约前主体评级、交易所、违约前债项评级、总资产周转率、货币资金、净资产增长率、经营活动现金流量占营收比、GDP、PPI、注册地、短期国债利率、宏观经济景气指数(先行指数)、债券类型和所属行业的特征重要度较高,在信用债违约中值得关注。该研究可为金融风险预测提供一种有效方法,对于投资者和金融机构的风险预警具有重要参考意义。

In order to enhance the accuracy and stability of credit bond default prediction for the purpose of financial risk management,a multi-objective optimized weighted soft voting ensemble algorithm is proposed by taking the credit bonds(January 1,2014,to December 31,2021)as the object of study.In this algorithm,the recognition capability of each base classifier is quantified by calculating their fuzzy densities,and the multi-objective particle swarm optimization algorithm is used to slove the weights of the base classifiers.In comparison with other individual classifiers such as support vector machine,logistic regression,Gaussian naive bayes,multi-layer perceptron(MLP),as well as other ensemble algorithms like voting and stacking,feature importance analysis is conducted by means of the expected permutation feature importance(PFI)algorithm.The results indicate that the weighted soft voting ensemble algorithm exhibits outstanding performance in credit bond default prediction.It not only enhances the performance of individual algorithms but also demonstrates higher accuracy,precision,and AUC values compared to other ensemble algorithms.Features with higher importance in credit bond default prediction include the issuer′s credit rating,exchange,bond rating prior to default,total asset turnover ratio,monetary funds,net asset growth rate,operating cash flow as a percentage of revenue,GDP,PPI,registered location,short-term government bond interest rates,leading economic indicators,bond type,and industry sector.This research can provide an effective approach for financial risk prediction,offering valuable insights for investors and financial institutions in the risk warning.