摘要
文章利用金融高频数据交易量信息,将深度学习长短期记忆(LSTM)模型应用于Va R风险管理。利用含有交易量信息的高频数据,结合LSTM模型,构造了LSTM-RV已实现波动率动态预测模型,利用半参数极值理论(EVT)方法估计收益率分位数,构建了LSTM-RV-EVT风险管理Va R模型。实证分析表明,相对于传统HAR(异质自回归)波动率预测模型,LSTM-RV预测模型的预测准确率显著提高,LSTM-RV-EVT模型比传统VaR模型和未利用交易量信息LSTM的VaR模型表现更好。
This paper applies the deep learning Long Short-Term Memory(LSTM)model to VaR risk management by using trading volume information of financial high-frequency data.And then,the paper is based on the high-frequency data containing trading volume information and combined with the LSTM model to construct thedynamic prediction model of LSTM-RVrealized volatility.Finally,the paper builds LSTM-RV-EVT risk management VaR model by using semi-parameter extreme value theory(EVT)method to estimate the quantile of return rate.The empirical analysis shows that,compared with the traditional HAR(Heterogenetic Auto Regressive)volatility prediction model,the prediction accuracy of the LSTM-RV model is significantly improved,and the LSTM-RV-EVT model performs better than the traditional VaR model and the VaR model without using trading volume information LSTM.
作者
刘广应
吴鸿超
孔新兵
Liu Guangying;Wu Hongchao;Kong Xinbing(School of Statistics and Mathematics,Nanjing Audit University,Nanjing 211815,China)
出处
《统计与决策》
CSSCI
北大核心
2021年第8期136-140,共5页
Statistics & Decision
基金
国家自然科学基金资助项目(71971118,11831008,11571250)
国家社会科学基金一般项目(19BTJ035)
江苏省自然科学基金面上项目(BK20181417)
江苏省高等学校自然科学研究重大项目(17KJA110001)
江苏省研究生科研与实践创新项目(KYCX18_1688)
