Financial Time Series Prediction Based on Deep Learning

Abstract

By combining wavelet analysis with Long Short-Term Memory (LSTM) neural network, this paper proposes a time series prediction model to capture the complex features such as non-linearity, non-stationary and sequence correlation of financial time series. The LSTM is then applied to the prediction of the daily closing price of the Shanghai Composite Index as well as the comparison of its prediction ability with machine learning models such as multi-layer perceptron, support vector machine and K-nearest neighbors. The empirical results show that the LSTM performs a better prediction effect, and it shows excellent effects on the static prediction and dynamic trend prediction of the financial time series, which indicates its applicability and effectiveness to the prediction of financial time series. At the same time, both wavelet decomposition and reconstruction of financial time series can improve the generalization ability of the LSTM prediction model and the prediction accuracy of long-term dynamic trend.