基于分解-重构-集成框架下的沪铜期货波动率预测  

Prediction of Shanghai Copper Futures Volatility Based on Decomposition Reconstruction Integration Framework

作　　者：景越 孙景云 Jing Yue;Sun Jingyun(Lanzhou University of Finance and Economics)

机构地区：[1]兰州财经大学

出　　处：《哈尔滨师范大学自然科学学报》2024年第6期34-43,共10页Natural Science Journal of Harbin Normal University

基　　金：国家自然科学基金(72061020);2022陇原青年创新创业人才项目;兰州财经大学“金融统计”学科科研融合团队

摘　　要：针对沪铜期货价格数据的复杂性和长期依赖性,基于“分解-重构-集成”思想,提出了一种新的波动率预测模型:CEEMDAN-SE-CNN-LSTM-SVR.首先,使用自适应噪声完备经验模态分解(CEEMDAN)将波动率分解为多个本征模态分量(IMF)与残差项,并依据样本熵(SE)值重构为高、低频分量,以降低序列建模的复杂度;其次,使用CNN-LSTM提取各分量的空间特征和时间信息并分别进行预测;然后,将各分量预测值利用支持向量回归(SVR)进行非线性集成,得到最终预测结果.实证结果表明:利用该模型预测沪铜期货波动率,其MAE相较于传统GARCH(1,1)模型减少了37%,其MAPE相较于SVR模型减少了70.1%,其RMSE相较于CNN-LSTM模型减少了22.5%.同时,对不同样本区间的波动率进行预测,验证了该模型对沪铜期货波动率预测的有效性和稳定性.In response to the complexity and long-term dependence of Shanghai copper futures price data,based on the idea of"decomposition reconstruction integration",a new volatility prediction model:CEEMDAN-SE-CNN-LSTM-SVR is proposed in this paper.Firstly,adaptive noise complete empirical mode decomposition(CEEMDAN)is used to decompose volatility into multiple intrinsic mode components(IMF)and residual terms,and reconstructed into high and low frequency components based on sample entropy(SE)values to reduce the complexity of sequence modeling.Secondly,use CNN-LSTM to extract spatial and temporal features of each component and make predictions separately.Then,the predicted values of each component are nonlinearly integrated using support vector regression(SVR)to obtain the final prediction result.The empirical results show that using the model proposed in this paper to predict the volatility of Shanghai copper futures,its MAE is reduced by 37%compared to the traditional GARCH(1,1)model,its MAPE is reduced by 70.1%compared to the SVR model,and its RMSE is reduced by 22.5%compared to the CNN-LSTM model.At the same time,predicting the volatility of different sample intervals validated the effectiveness and stability of the model for predicting the volatility of Shanghai copper futures.

关 键 词：沪铜期货 自适应噪声完备经验模态分解 CNN-LSTM 支持向量回归 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]