Stock movement prediction: A multi‐input LSTM approach

Abstract Generally, the nonlinear and non‐stationary financial time series becomes an obstacle in the process of stock movement prediction. But the recent theories of machine learning and deep learning have provided with some new solutions. Based on LSTM (long short‐term memory), we propose a hybrid model of wavelet transform (WT) and multi‐input LSTM to predict the trend of SSE composite index. It can mine valid data in time series and support different types of data as input. The whole model is divided into two stages. In the first stage, we adopt the level 1 decomposition with db4 mother wavelet to eliminate noise. In the second stage, combinative and qualitative analysis was made base on the data from Chinese stock market, US stock market, and technical indicators as input. According to the result, the proposed model, with the accuracy of 72.19%, performs better than single‐input LSTM, decision tree, random forest, Support Vector Machine (SVM), and XGBoost.