基于统计模型和机器学习模型的金融时间序列预测

Financial Time Series Forecasting Based on Statistical Models and Machine Learning Models

摘要: 金融时间序列的预测一直是投资者关注的重点，针对金融时间序列预测难度大、准确度低、影响因素众多等问题，以深证成指、上证指数和中证500的收盘价为例，分别构建自回归求和移动平均(ARIMA)模型、卷积神经网络(CNN)模型、长短期记忆神经网络(LSTM)模型。CNN-LSTM模型和CNN-LSTM-ARMA模型用于这三种金融时间序列的收盘价预测的研究。通过比较，发现CNN-LSTM-ARMA模型比单一预测模型和CNN-LSTM模型预测收盘价的准确性高，拟合效果好。

Abstract: The prediction of financial time series has always been the focus of investors’ attention, in view of the problems of difficult forecasting, low accuracy and many influencing factors of financial time series, taking the closing prices of Shenzhen Stock Exchange Component Index, Shanghai Composite Index and CSI 500 as examples, the Autoregressive Sum Moving Average (ARIMA) model, Convolutional Neural Network (CNN) model, Long Short-Term Memory Neural Network (LSTM) model are constructed respectively. The CNN-LSTM model and the CNN-LSTM-ARMA model are used for the study of closing price forecasting for these three financial time series. Through comparison, it is found that the CNN-LSTM-ARMA model has higher accuracy and better fitting effect than the single prediction model and the CNN-LSTM model in predicting the closing price.