分位数惩罚整合模型及其在商业银行系统性风险中的应用

Quantile Penalized Integrative Method and Its Application in the Systemic Risk of Commercial Banks

大数据通常由不同来源的数据组合而成,且具有高维特征,挖掘不同来源数据间的异质性和关联性并降维是亟需解决的问题.基于此,文章提出了分位数惩罚整合模型,并给出其模型表示和模型算法.该模型既可以对不同来源数据进行建模和变量选择,又同时考虑了不同来源数据间的异质性和关联性.数值模拟结果表明:分位数惩罚整合模型在预测性能和变量选择方面都具有明显的优势.此外,将该模型应用于商业银行系统性风险测度中发现,分位数惩罚整合模型在实际应用中也有较好的表现.

Big data usually combines data from different sources,and has high dimensional characteristic.It is a problem that how to mining the heterogeneity and correlations of different datasets and achieves dimension reduction.Hence,this paper proposes quantile penalized integrative method.A set of modeling methodology,includes mathematics expression,parameter estimation,is studied in detail.It can not only simultaneously build up models and select variables using data from different sources,but also considering the heterogeneity and correlations of data from different sources.The numerical simulation results show that the quantile penalized integrative method has a significant advantage in both model prediction and variable selection.Finally,the quantile penalized integrative method is applied to the measurement of the systemic risk of commercial Banks.The results show that the quantile penalized integrative method has a good performance in practical application.