基于分解重构的欧盟碳排放权市场波动率预测研究——新冠疫情、俄乌冲突背景下  

Research on Predicting the Volatility of EU Carbon Emission Rights Market Based on Decomposition and Reconstruction-under the Background of COVID-19 and Russia-Ukraine Conflict

作　　者：靳慧娜 张金良[1] 白祥 JIN Huina;ZHANG Jiniang;BAI Xiang(School of Mathematics and Statistics of Henan University of Science and Technology)

机构地区：[1]河南科技大学数学与统计学院

出　　处：《上海节能》2024年第4期630-640,共11页Shanghai Energy Saving

基　　金：国家自然科学基金(51675161)。

摘　　要：碳排放权交易系统受多种因素影响具有强非线性、强波动性等特点,碳排放权收益波动率的预测极具挑战性。近年来突发的新冠疫情和俄乌冲突对碳市场带来了前所未有的冲击。以GARCH(1,1)波动率作为欧盟碳排放权的“真实”波动率,针对近几年动荡不安的碳市场,提出了一种基于自适应噪声完备集合经验模态分解(CEEMDAN)、样本熵(SE)、支持向量回归(SVR)、长短时记忆神经网络(LSTM)的波动率复合预测模型,即CEEMDAN-SE-SVR/LSTM/LSTM。该模型通过CEEMDAN和SE捕捉碳排放权波动率不同时间尺度上的特征,利用传统机器学习SVR在小样本上的鲁棒性以及深度学习LSTM模型长记忆特征对波动率实现精准预测。以近期欧盟EUA期货数据为样本进行了实证分析,结果表明,CEEMDAN-SE-SVR/LSTM/LSTM模型预测精度和鲁棒性优于其它参考模型。The carbon emission trading system is influenced by various factors and has characteristics such as strong non-linearity and strong volatility.Predicting the volatility of carbon emission trading returns is extremely challenging.The sudden COVID-19 and Russia-Ukraine conflict in recent years have brought unprecedented impact on the carbon market.A volatility composite prediction model based on Adaptive Noise Complete Set Empirical ModeDecomposition(CEEMDAN),Sample Entropy(SE),SupportVectorRegression(SVR),and Long Short Term Memory Neural Network(LSTM)is proposed for the turbulent carbon market in recent years,using GARCH(1,1)volatility as the"real"volatility of EU carbon emission rights.This model is CEEM-DAN-SE-SVR/LSTM/LSTM.This model captures the characteristics of carbon emission rights volatility at dif-ferent time scales through CEEMDAN and SE,and utilizes the robustness of traditional machine learning SVR on small samples and the long memory features of deep learning LSTM models to achieve accurate prediction of volatility.Empirical analysis was conducted using recent EU EUA futures data as a sample,and the results showed that the CEEMDAN-SE-SVR/LSTM model had better prediction accuracy and robustness than other referencemodels.

关 键 词：碳排放权波动率 CEEMDAN 样本熵 LSTM SVR 

分 类 号：F831.5[经济管理—金融学] X196[环境科学与工程—环境科学] F224