The Analysis of Time Series with Use of Modern Information Technologies

Forecasting oil prices represents a topical task of global scale. In conditions of large flow of heterogeneous information, there is a need to use effective methods for processing them in order to develop operational management decisions. In particular, machine learning methods have become increasingly popular in recent times, which often turn out to be more accurate than traditional approaches. The aim of the paper is to compare econometric models with machine learning algorithms, such as gradient boosting and a long short-term memory (LSTM) neural network, to determine the best model in terms of the oil prices prediction. Information base in this study is represented by the monthly reports of international organizations, software product Thomson Reuters, as well as a number of other resources. In total, the collected database had 84 factors for the period from 01/2002 to 09/2017. Also articles directly related to the oil market were collected. On their basis, a text analysis was conducted in order to obtain more accurate forecasts. All the calculations were performed by using machine learning libraries in Python and R. As a result, gradient boosting model, ARIMA, and their ensemble achieve the best quality. Possible areas for further research include the creation of more integrated models that take into account investor sentiment, as well as the development of advanced hybrid models.