Stock Price Prediction using LSTM model

Abstract

Accurate forecasting of stock prices is becoming increasingly important in stock markets where returns and risks are highly volatile. The paper compares the accuracy of conventional time series forecasting methods and LSTM-based neural networks based on S&P data from 2012 to 2017 using python and Keras. The impact of the method level and network structure on the results is further investigated by comparing the indicators. The results display that the autoregression model and one-layer LSTM network show the most satisfying performance of metrics for S&P index prediction. The mean absolute percentage error of these two methods approximates 0.01. Finally, compare and evaluate the possible errors caused by various forecasting methods to better exploit the positive effects of traditional learning methods and neural network methods on stock price forecasting.