Abstract: Aiming at the problem that the existing prediction modeling methods are difficult to efficiently extract the complex change rules of intraday trading volume distribution, which affects the implementation effect of VWAP strategy, this paper proposes a forecasting method M-LSTM of intraday trading volume distribution of stock index based on LSTM-Attention under MEMD decomposition.Firstly, the time series of interval multidimensional trading volume are decomposed into several independent IMF synchronously using MEMD. Secondly, the high-frequency IMF in each dimension decompositionis de-noised and reconstructed, and the intradaytrading volume distribution prediction model based on LSTM-Attention neural network is built, and then the effectiveness of the prediction model is deeply analyzed under different trend stages of stock indexes. Finally, M-LSTM, ARIMA, SVR and other mainstream methods are used to forecast the intraday trading volume distribution of four representative stock indexes such as Shanghai Composite Index. The experimental results show that M-LSTM having smaller prediction erroris a more effective method for predicting intraday trading volume distribution of stock indexes.

Key words: intraday trading volume distribution, VWAP strategy, MEMD, LSTM-Attention

摘要： 针对现有预测建模方法难以高效提取日内交易量分布复杂变化规律,影响VWAP策略执行效果的问题,本文提出一种MEMD分解下基于LSTM-Attention的股市指数日内交易量分布预测方法M-LSTM。首先,运用MEMD方法将区间多维交易量时序同步分解为若干个独立的本征模态函数(IMF);其次,对各维度分解中高频IMF进行去噪与重构,构建基于LSTM-Attention神经网络的日内交易量分布预测模型,并深入分析股票指数不同走势阶段下模型预测的有效性;最后,分别采用M-LSTM、ARIMA以及SVR等主流方法,对上证指数等四个代表性指数的日内交易量分布进行预测。实验结果表明:M-LSTM预测误差更小,是一种更有效的股票指数日内交易量分布预测方法。

关键词: 日内交易量分布, VWAP策略, 多元经验模态分解, LSTM-Attention