摘要
运用贡献度随机森林方法(CRF)方法探讨公司债财务指标比率与其违约率的关系.运用连续属性离散化方法(OB)进行财务指标最优降维;运用WOE变换进行模型变量约简.研究表明,CRF模型的分类性能显著优于其他模型,测试集评估总体正确率达90.47%,AUC统计量、AR比率及K-S值分别提升了2.6%、7.6%、4.38%,变量贡献度量化了各财务指标对违约率影响,为诠释随机森林预测机制提供了依据.
The contribution forest model（CRF） was used to research the inner connection between the corporate bonds and its financial index ratio,. The method of discretization and WOE transformation were applied to reduce the dimension of these indexes. The results show that the CRF model＇s performance significantly outperforms the other models, and the per- formance of the model on test dataset reaches a accuracy of 90.47%. And the other assessment indexes,AUC statistics, AR ra- tio and K-S values, are improved by 2.6%, 7.6%, 4. 38%. Furthermore, the contribution of variables evaluated its influence on probability of default in a quantitative way, which provides a new point of view to interpret the process of forecast of random forest.
作者
汪政元
伍业锋
WANG Zheng-yuan WU Ye-feng(School of Economics, Jinan University, Guangzhou 510632)
出处
《经济数学》
2016年第3期33-40,共8页
Journal of Quantitative Economics
基金
中央高校基本科研业务费专项资金暨南远航计划(12JNYH002)
关键词
财务管理
违约预测
实证分析
贡献度随机森林
连续属性离散化
WOE变换
financial management
default prediction
empirical analysis
contribution andom forest model
discretiza-tion
WOE transformation
