An Empirical Analysis for Forecasting Stock Index based on LSTM Neural Network

In this paper, we study a stock price forecasting model based on a trained neural network for the Chinese stock market. We propose a Long Short-Term Memory (LSTM) network model to predict the closing price on a targeted day. The transaction data of four representative stock indices are investigated in the empirical analysis, including Shanghai Composite Index (stock code 000001), Shenzhen Composite Index (stock code 399001), CSI 300 Index (stock code 399300) and SSE 50 Index (stock code 000016). To predict the closing price for the next period, five important financial characteristics of the transaction data are selected as the input features, such as the opening price, historical closing price, trading volume, highest price, and lowest price. In addition, the proposed LSTM model is compared with the traditional Recurrent Neural Network (RNN) model on four performance measures. In numerical results, we indicate the applicability of the proposed method through the empirical analysis of those four stock indexes. Our works could contribute to an effective guidance in practical trading analysis for investors when making rational investments in the stock markets.