An Automated Procedure Of Indentifying SARIMA Models For Macroeconomic Time Series

Abstract

An automated procedure of building the best SARIMA model for a given time series is developed and implemented. Adjusted mean squared error of one-step-ahead forecast, mean absolute percentage value and mean relative range of confidence intervals are chosen as selec¬tion criteria. Short-term and mid-term forecasts of some indicators of the commodity-producing sector of Belorussian economy are built.