非均衡数据的债券违约预警研究

将上海交易所和深证交易所发行的30只违约债券和468只未违约债券作为研究样本，将债券是否违约设定为一个二分类问题进行识别分析，针对该问题构建了基于SVM的ADmR-AdaboostSVM分类模型；从企业资本结构、盈利能力、现金流量、偿债能力4个评估因素中筛选16个预警指标，运用ADASYN方法进行过采样合成新样本点，将特征提取mRMR方法引入债券违约领域，得出长期负债率、资本收益率、成本费用利润率以及股权比例这4个变量作为债券违约的最终预警指标，在此基础上运用AdaboostSVM模型进行风险识别。研究结果表明：在建模过程中克服了样本非均衡化问题使得分类精度显著提高，同时通过解决高维数据冗余问题，识别违约债券的准确率进一步提高，反复验证表明该模型具有较强的稳健性和有效性，具有一定的应用价值。

Research on Early Warning of Bonds Default Based on Unbalanced Data

30 default bonds and 468 non-default samples were selected as research samples from Shanghai Stock Exchange and Shenzhen Stock Exchange, whether the bonds were defaulted was set as a binary classification problem for identification and analysis, and ADmR-AdaboostSVM classification model based on SVM was constructed for this problem. This article selects 16 early warning indicators from such four evaluation factors as enterprise capital structure, profitability, cash flow, and solvency, uses ADASYN method for oversampling and synthesizing new sample points, introduces mRMR method of feature extraction to bonds default field to obtain such four variables as long-term debt ratio, the rate of return on capital, profit margin on costs, and equity ratio as final early warning indicators of bonds default, and on this basis, uses AdaboostSVM model to conduct risk identification. Research results show that the sample unbalance problem is overcome during modeling process to make classification accuracy significantly improved, at the same time, the accuracy for identifying defaulted bonds is further achieved by solving the problem of high-dimensional data redundancy. Repeated verifications show that this model has strong robustness and effectiveness and has certain application value.