基于混频数据驱动神经网络模型的波动率预测研究  

Research on volatility prediction based on mixed frequency data-driven neural network model

作　　者：汪刘凯 张小波 闫相斌 王未卿[1] WANG Liukai;ZHANG Xiaobo;YAN Xiangbin;WANG Weiqing(The School of Economics and Management,University of Science and Technology Beijing,Beijing 100083,China;School of Business,Guangdong University of Foreign Studies,Guangzhou 510420,China)

机构地区：[1]北京科技大学经济管理学院,北京100083 [2]广东外语外贸大学商学院,广州510420

出　　处：《系统工程理论与实践》2023年第12期3488-3504,共17页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(72025101,72301025,71729001);中国博士后科学基金(2021M700380);中央高校基本科研业务经费(FRF-TP-22-060A1,FRF-BR-23-08B);北京市社会科学基金(23GLB022)。

摘　　要：金融市场价值波动与经济政策等宏观环境密切相关,其影响要素多源,价值波动往往表现出复杂的统计特征与变化规律,现有的波动率预测模型难以有效地预测其价值波动规律.面对价值波动中风险要素多源、频率多样、关系非线性的潜在挑战,考虑到深度学习框架下CNN和LSTM的计算优势,本文提出了基于混频数据反向抽样的CNN-LSTM的波动预测模型:MDNN (mixed frequency data-driven neural network),该模型既有效提取多元时序数据的时空特征、又充分利用混频信息,使其预测能力与泛化能力得到有效提升.选取常见的供应链金融质押物铜、铝和锌作为研究对象,样本外预测结果表明:相比于基准模型,MDNN更加准确、有效地预测出质押物已实现波动率,其稳健性检验也表明实证结论的可靠性.Financial market value fluctuation is closely related to economic policies and other macro-environments,which have multiple sources of influence factors,and its value fluctuations often shows complex statistical characteristics and changes.The existing models are difficult to predict the value fluctuation effectively.Facing the potential challenge of multi-source,diverse frequency,and nonlinear relationship in the change of pledge value,and considering the computational advantages of CNN and LSTM under the framework of deep learning,this paper proposes a novel volatility forecasting model of CNN-LSTM based on mixed frequency data:Mixed frequency data-driven neural network(MDNN),which not only effectively extracts the spatio-temporal characteristics of multivariate time series data,but also makes full use of mixed frequency information.The prediction ability and generalization ability of this approach are improved.To illustrate the efficacy of our method,empirical studies on Copper,Aluminium,and Zinc.The out-of-sample prediction results show that:Compared with the benchmark models,MDNN predicts the pledge realized volatility more accurately and effectively.And robustness test also shows the reliability of the empirical conclusion.

关 键 词：已实现波动率 混频数据 深度学习 MDNN 

分 类 号：F224.0[经济管理—国民经济]