基于LSTM等深度学习方法的股指预测研究

Research on Stock Index Prediction Based on Deep Learning Methods Such as LSTM

为了改善传统时间序列方法无法在预测模型中添加相关变量等缺点,并提高股指预测精度,运用LSTM神经网络等深度学习方法对我国上证指数及沪深300指数进行预测分析,并将预测结果与RNN、CNN、ARMA等模型进行比较,然后在模型中加入百度指数测试其对预测精度的影响,最后检验LSTM模型对训练步长的敏感性。研究结果表明,LSTM能够实现对股指的精准预测,其预测评价指标MAE、MAPE、RMSE分别为0.008、0.025、0.011,预测误差低于其它模型,加入百度指数可进一步提升其预测能力,但改变LSTM模型训练步长对结果影响不大。因此,LSTM模型在金融经济预测领域有较高的应用价值。

In order to improve the shortcomings such as impossible to add relevant variables to the prediction model of traditional time series methods and improve the accuracy of stock index prediction, this paper uses the deep learning method (LSTM neural network) to predict and analyze China's Shanghai Composite Index and the Shanghai and Shenzhen 300 Index. The prediction results are compared with RNN, CNN, ARMA and other models. Then, Baidu index is added to the model to test its influence on prediction accuracy. Finally, the sensitivity of LSTM model to train step size is tested. The research results show that LSTM can accurately predict the stock index. The forecasting indicators MAE, MAPE and RMSE are 0.008, 0.025 and 0.011 respectively. The prediction error of LSTM is lower than other models and adding Baidu index data can further improve the forecasting ability. Therefore, the LSTM model has application value in the field of financial economic forecasting.