Financial Time Series Forecasting on Base of the Deep Learning Networks

Financial time series forecasting using past observations is a subject of considerable interest among scientists, entrepreneurs, economists, financiers, investors, and traders. This task is complicated by the fact that to obtain an accurate forecast, it is necessary not only to have a large amount of information, but also to be able to interpret it correctly, which is often impossible. One way to predict time series is to model them based on past values, followed by extrapolation. The scientific literature describes many ways to model time series, some of which are based on deep learning networks. There are several deep learning network architectures that can be used in the task of forecasting time series: multilayer perceptron, convolutional neural networks, recurrent neural networks. The goal of this paper is to compare the effectiveness of these architectures in the task of forecasting financial time series.