Research on the S&P 500 Index Based on LSTM and GRU

Abstract

Forecasts regarding the prices of stock indices are commonly observed in both academic and corporate circles. Exponential prices are difficult to predict because of uncertain noise. Due to advancements in computer science, neural network has been applied in various industrial fields. This article examines three models, Long Short Term Memory (LSTM) networks and Gate Recurrent Unit (GRU), and their combination. At the same time, it’s also investigated that how the performance of the models will change when incorporating technical indicators such as the weighted moving average and Bollinger bands and macroeconomic indicators which include CBOE Volatility Index and Effective Federal Funds Rate into analysis. The experimental findings indicate that LSTM performs best in the experiment of predicting S&P500 index, and adding indicators to the model does not make the model more effective. This article provides more perspectives on the application of machine learning in stock price prediction, and provides relevant theoretical guidance for the selection of variables in stock price prediction practice.