Intelligent Methods for Predicting Financial Time Series

Abstract

In modern conditions, to ensure the successful implementation of any activity, it is necessary to perform high-quality and efficient forecasting of current processes. The scope of application is expanding, making the task of forecasting even more important and complex. The increasing role of forecasting in the modern world has given rise to over a hundred models and methods of forecasting. For this reason, the challenge becomes to select the optimal variant of forecasting the process or system under study. In the present article, the main mathematical methods of forecasting time series are analyzed and their advantages and disadvantages described. Criteria for the accuracy of forecasting models are defined. Practical application of various models is considered. The possibilities for implementing forecasting models are investigated. The capabilities of the Python programming language for developing forecasting models are evaluated. The paper solves the problem of constructing a weighted-average forecast, which consists of several individual forecasts. Original forecasting models that were used in the combination included Arima, gradient boosting and a fully connected feed-forward neural network. Neural networks are growing more relevant today, as they enable forecasting in the event of a crisis and uncertainty. During the implementation of the programming solution, the mean absolute errors were computed for each forecasting method as well as for the weighted-average forecast. #COMESYSO1120.