基于机器学习和经验模态分解的跨期套利研究  

Research on Intertemporal Arbitrage Based on Machine Learning and Empirical Mode Decomposition

作　　者：周亮 陈辰 李宁 ZHOU Liang;CHEN Chen;LI Ning(School of Finance,Hunan University of Finance and Economics,Changsha 410205,China;School of Finance,Southwestern University of Finance and Economic,Chengdu 611130,China)

机构地区：[1]湖南财政经济学院财政金融学院,长沙410205 [2]西南财经大学金融学院,成都611130

出　　处：《西南大学学报（自然科学版）》2022年第1期148-159,共12页Journal of Southwest University(Natural Science Edition)

基　　金：国家社会科学基金项目(20BJL061);湖南省教育厅科学研究项目(21B0839).

摘　　要：采用滚动经验模态分解(EMD)方法对沪深300股指期货当月和下月合约的价差波动进行分解,分别利用Elman网络、随机森林(RF)、支持向量回归(SVM)3种机器学习模型及自回归移动平均模型(ARIMA)对不同频率信号进行分析,合成最终的预测结果,并根据预测结果设计跨期套利策略.研究结果表明:SVM,RF和ARIMA模型的预测精确度相对Elman网络较高,所有模型均能取得较高的套利收益,将非线性模型和线性模型融合使用能够改善模型的风险控制能力;将机器学习预测与EMD分解技术相融合可以在不提高风险的同时大幅度提高模型的收益率,从而使得模型的夏普比率和索提诺比率均有较大幅度上涨;分样本检验、全IMF信号预测以及基于商品期货市场的套利分析,均证明融合EMD的机器学习模型可以获得比纯机器学习模型更优异的套利效果.研究结论有助于促进人工智能与金融学的交叉融合研究,同时也为期货投资提供了理论和现实参考.This paper used rolling EMD(Empirical Mode Decomposition)method to decompose the price gap of the CSI 300 stock index futures contract of the current month and the next month,and used three machine learning models(Elman network,RF,SVM)and ARIMA model to analyze and synthesize signals of different frequencies,and designed intertemporal arbitrage strategies based on the forecast results.The research results show that:the prediction accuracy of SVM,RF and ARIMA models is higher than that of Elman network.All models can achieve higher arbitrage returns,and the use of model fusion which combines liner and nonliner models can improve the risk control ability of the model.The combination of machine learning prediction and EMD decomposition technology can greatly increase the profitability of the model without increasing the risk,so that the Sharpe ratio and the Sotino ratio of the model are both larger.Sub-sample test,full IMF signal prediction and arbitrage analysis based on the commodity futures market have all proved that the machine learning model integrated with EMD can achieve better arbitrage effects than pure machine learning models.The research conclusions help to promote the cross-integration research of artificial intelligence and finance,and also provide theoretical and practical references for futures investment.

关 键 词：机器学习 经验模态分解 跨期套利 期货投资 人工智能 

分 类 号：F830.9[经济管理—金融学]