The Application Of Deep Learning To Shanghai Composite Index Return Prediction

The Shanghai Composite Index is currently the most representative of the domestic securities market information.Its year-on-year logarithmic return rate has the statistical characteristics of general financial time series.It is suitable to use the deep learning time series model to analyze the law of its historical evolution and predict the future trend,which is of great significance for risks control and trading strategies.This article describes the cycle analysis and trend analysis commonly used in the technical analysis of the securities market with the deep learning method.First,the spectral characteristics of the rate of return are analyzed by the spectral analysis method,the main energy points of the spectrum are found,the sine period is used to fit the main frequency period,a composite periodic function is constructed under the framework of deep learning,and the gradient descent method is used to fit the period parameters such as amplitude,phase,etc.,and verify that the Shanghai Composite Index's return rate has typical economic cycle characteristics.Secondly,the linear models ARMA and GARCH are used to analyze the trend and volatility of the return rate,which build a benchmark for the subsequent deep learning modeling.Use the neural network in the deep learning field to manually implement the ARMA model,then use the LSTM model commonly used in time series analysis to analyze the trend of return,and then build a Deep AR model based on the LSTM model to analyze the trend and volatility of return.Finally,the ARMA model and the Deep AR model are fused,which combines the advantages of the ARMA model and the Deep AR model.