基于深度学习的上证综指波动率预测效果比较研究  

Forecasting Volatility of Shanghai Composite Index with Deep Learning

作　　者：陈卫华[1] CHEN Wei hua(School of Statistics and Management, Shanghai University of Finance and Economics, Shanghai 200433, Chin)

机构地区：[1]上海财经大学统计与管理学院,上海200433

出　　处：《统计与信息论坛》2018年第5期99-106,共8页Journal of Statistics and Information

基　　金：国家社会科学基金重点项目<我国创新驱动转型发展评价指数的构建与应用研究>(16ATJ004)

摘　　要：在高频波动率预测领域,首次运用深度学习对波动率进行样本外预测,以提高波动率预测精度,并将预测结果与19种经典模型作对比以评价预测效果。研究发现:深度学习在5种损失函数下预测精度都排第1。与排名第2的对比模型相比,预测精度在不同损失函数下最大提升13.16%,最小提升9.72%。最后,深度学习受关键参数历史天数变化影响较小,在大多数历史天数下,LSTM模型在检验模型中预测效果依旧最好,而且随着历史天数的增加,模型的预测效果趋于稳定。This paper introduces the deep learning method to predict the high-frequency volatility of stock market and tries to improve the accuracy of prediction.To test the predictive effect of LSTM model,the paper compares the LSTM model with other 19 kinds of classic models that widely used to predict the volatility in stock market.The empirical results show that the LSTM model ranks the first under all five loss functions.As the LSTM model ranks the first,the prediction accuracy increases from 9.72% to13.16% under different loss functions compared with other 19 comparable models.Finally,as historical days changing,the prediction performance of the LSTM model is less affected,and the prediction performance of the LSTM model tends to be more stable with the increase of historical days.

关 键 词：深度学习 LSTM模型 已实现波动率模型 

分 类 号：F830[经济管理—金融学]