摘要
基于线性分位数回归和三种非线性分位数回归模型,对我国14家上市商业银行的股价收益率VaR与风险状态下银行的系统风险CoVaR进行了估算。通过6种回测法的8个检验统计量对模型进行了综合评估,并基于单一模型构建了更加综合的组合模型。研究表明:就准确性和稳健性而言,不同机器学习分位数回归模型各有所长;线性分位数回归模型较综合,可应用情景多;由线性分位数回归、神经网络分位数回归和随机森林分位数回归构建的组合模型比单一模型更加准确有效,综合有效性最强。
Based on the linear quantile regression method and three types of nonlinear quantile regression models,the paper estimated the share price yields VaR and the systemic risk CoVaR in risk state of 14 listed commercial banks.In this paper,the research applied 8 testing statistical magnitudes of 6 backtesting methods to make a comprehensive assessment,and then constructed a more integrated and combinatorial model based on the single model.The results showed that,considering accuracy and robustness,each machine learning quantile regression method had its own advantages.And linear quantile regression method was more comprehensive,which was suitable for multiple locations.Besides,the comprehensive model,combining the methods of linear quantile regression,neural network quantile regression and random forest quantile regression,was more accurate and effective than a single model.
作者
王周伟
雷潇
WANG Zhou-wei;LEI Xiao(School of Finance and Business,Shanghai Normal University,Shanghai 200234,China)
出处
《统计学报》
2022年第4期81-94,共14页
Journal of Statistics
基金
国家自然科学基金面上项目(71973098)。
关键词
商业银行
条件风险价值
机器学习分位数回归
回测检验
分位数回归组合预测
commercial bank
conditional value at risk
machine learning quantile regression
backtesting inspection
combined estimation of quantile regression
