摘要：In forecasting aggregated time series different approaches can be used. The aim of this paper is to analyze and forecast aggregated time series of the total energy consumption in Croatia up to 2014, using Box-Jenkins methodology and applying two different approaches: direct and indirect. Forecast values were calculated indirectly by aggregating the forecast values which are determinate (with the different representative forecasting models) for all variables components in an aggregated time series. The components of total energy consumption in Croatia are: consumption of coal, fuel wood, liquid fuel, natural gas, hydro power and electricity. Direct approach in forecasting means that forecast values are determined directly applying appropriate forecasting model on aggregated time series. In forecasting Croatia’s total energy consumption, ARIMA (1,1,1) model was selected as representative. The same model is chosen for dominant variable component: liquid fuel. The differences in forecast values of total energy consumption calculated directly or indirectly, are the consequence of applying different (representative) forecasting models for the different time series components of total energy consumption. From the statistical point of view, direct approach is preferred, but in a compliance with the expectation of continuing recession period in Croatia in the next two years, forecast values of total energy consumption calculated by indirect approach are more realistic then the same calculated using direct approach.
关键词：forecasting; aggregated time series; total energy consumption; stationarity; Box-Jenkins approach; ARIMA model; ADF test.
