In this study, a novel hybrid model for share price index futures forecasting named WT-ARIMA-LSTM is proposed. In this hybrid model, share price index futures are decomposed to extract data characteristics at different time scales by the wavelet transform and the ARIMA-LSTM model are applied to predict the close price of futures. The findings of the study are as follows. 1) The DWT hybrid model and the MODWT hybrid model have higher forecasting accuracy than some commonly used forecasting models under the three metrics of MAE, MAPE and RMSE. The DWT-ARIMA-LSTM model has better forecasting performance when the forecasting performance in different markets and the operational efficiency of the method are combined. 2) The DWT method is more applicable than the MODWT method in forecasting models of futures closing price series; the approximate signals obtained from the DWT decomposition have lower volatility and can better characterise the original signals. 3) The LSTM model has better prediction performance for noisy residual series, while the ARIMA model has better prediction performance for less noisy approximate signals. 4) Based on the forecasting results, a timing trading strategy is constructed that can maintain a robust return performance under different market conditions, especially on the risk side with significant advantages. In addition, this work examines the impact of the unexpected event of the COVID-19 epidemic on the forecasting performance of the model, and the results show that the model can adapt to different data structures to achieve more robust forecasting performance. This work provides insights into the integration of deep learning methods with econometric methods in the field of asset pricing.

在本研究中，提出了一种新颖的股指期货预测混合模型，名为 WT-ARIMA-LSTM。在该混合模型中，通过小波变换对股指期货进行分解，提取不同时间尺度的数据特征，并应用ARIMA-LSTM模型来预测期货收盘价。研究结果如下。1）DWT混合模型和MODWT混合模型在MAE、MAPE和RMSE三个指标下比一些常用的预测模型具有更高的预测精度。当结合不同市场的预测性能和方法的运行效率时，DWT-ARIMA-LSTM模型具有更好的预测性能。2）DWT方法比MODWT方法更适用于期货收盘价序列的预测模型；DWT分解得到的近似信号波动性较低，能够更好地表征原始信号。3）LSTM模型对于有噪声的残差序列有更好的预测性能，而ARIMA模型对于噪声较小的近似信号有更好的预测性能。4）根据预测结果构建择时交易策略，在不同市场条件下都能保持稳健的收益表现，尤其在风险面具有显着优势。此外，这项工作还考察了COVID-19疫情突发事件对模型预测性能的影响，结果表明该模型可以适应不同的数据结构以实现更稳健的预测性能。这项工作为资产定价领域的深度学习方法与计量经济学方法的整合提供了见解。