基于GARCH-MIDAS-dLSTM模型的棉花期货价格趋势拟合研究

Research on cotton futures price trend fitting based on GARCH⁃MIDAS⁃dLSTM model

利用机器学习对期货价格的趋势进行拟合研究的时候,在价格波动剧烈的情况下效果往往并不理想.期货基本面方面的因子大多都是月度或者年度之类的低频数据,不能与日度价格数据进行很好地融合,因此无法充分利用这类信息进行拟合研究.针对上述问题,提出一种基于混频数据的GARCH-MIDAS-dLSTM模型,以棉花期货作为研究标的,将多因子GARCH-MIDAS模型与双层LSTM模型进行融合并对棉花期货收盘价进行趋势拟合.实验结果表明,引入棉花期货基本面低频因子的GARCH-MIDAS-dLSTM模型MAE值为0.127 3,较引入之前提升了0.018 4,表明该模型能够在价格波动剧烈的情况下提高拟合结果的准确度并减小误差.

When using machine learning to fit the trend of futures price,the effect is not always ideal when the price fluctuates violently.Most of the factors in the fundamental aspects of futures are monthly or annually low⁃frequency data,which cannot be well integrated with the daily price data.In view of the above problems,this paper proposed a GARCH⁃MIDAS⁃dLSTM model based on mixing data,which used cotton futures as the research subject.The model integrated the multi⁃factor GARCH⁃MIDAS model with double⁃LSTM model and fit the trend of cotton futures closing price.The results showed that the MAE value of the improved model was 0.1273,which was 0.0184 higher than that of the original one,and indicated that the improved model could effectively promote the accuracy of the fitting results and reduce the error in the case of sharp price fluctuations.