Application of neural networks in the problem of forecasting financial time series

The purpose of the article is to predict future stock market prices using neural network models with the subsequent selection of the most accurate model. The article discusses the main types of neural networks used for time series analysis - long-term short-term memory network (LSTM) and managed recurrent block (GRU). These algorithms are types of recurrent neural networks (RNN). The algorithms were developed in the Google Colab environment in the Python programming language version 3.7 using the Pandas, NumPy, Scikit-learn, Statsmodels, Keras, Matplotlib libraries. The Russian stock market was selected for the study from 06/02/2014 to 11/16/2019. The sample included securities traded on the Moscow Stock Exchange, including shares of such companies as PJSC Aeroflot (AFLT), AK Alrosa (ALRS), PJSC Gazprom (GAZP), MMC Norilsk Nickel (GMKN), PJSC Severstal (CHMF). The results of the study led to the conclusion that the neural network configuration of LSTM is the most effective when creating a predictive model of artificial intelligence. The findings of the study also indicate that the use of tools based on artificial intelligence is an effective way to predict financial time series.