The article presents the results of experiments to study the possibilities of short-term forecasting of financial time series using various types of forecasting algorithms. There are results of testing of ARIMA method and Long short-term memory algorithm on a data set wich describes the value of the US dollar relative to the Russian ruble for one day, to predict the future value of the indicator under study. The estimation of the accuracy of forecasting of each of the methods and their suitability for solving such problems is made.

Keywords: forecasting, time series, neural networks, finance, regression algorithms, data analysis, reccurent neural networks, python, numpy, pandas, keras