Recurring neural networks applied to price forecasting in the stock market 2018-2024

Abstract

Predicting the direction of stock price trends is vital to optimally developing strategies for transactions that occur in the stock markets. Due to the risk and variable returns of the stock market, its prediction is a very important issue for those who invest in it. Having the ability to forecast the trend or price of stocks is very valuable information for investors. In the present work, some models have been proposed that may become viable for the prediction of stock market indicators, such as the case of the S&P 500. For this, the study of two different models has been carried out, which belong to the family of recurrent neural networks (LSTM and GRU). Finally, a comparison is made between the different models using the MSE (mean square error) metric. It is concluded that both models are suitable for predicting these stock indices, evidencing the great predictive power that artificial neural networks have. Black box models can be very advantageous, given that their parameters can be further optimized and thus way to obtain better results in predictions.