Description:
The primary goal of this paper is to describe several models that are currently used at the National Bank of the Republic of Macedonia (NBRM) for short-term forecasting of inflation - Autoregressive integrated moving average models (aggregated and disaggregated approach), three equation structural model and a dynamic factor model. Additionally, we evaluate models' out-of-sample forecasting performance for the period 2012 q3 to 2016 q2 by using a number of forecast evaluation criteria such as the Root Mean Squared Error, the Mean Absolute Error, the Mean Absolute Percentage Error and the Theil's U Statistics. Additionally, we constructed several composite forecasts in order to test whether a combination forecast is superior to individual models' forecasts. Our results point to three important conclusions. First, the forecasting accuracy of the models is highest when they are used for forecasting one quarter ahead i.e. the errors increase as the forecasting horizon increases. Second, the disaggregated ARIMA model has the smallest forecasting errors. Third, majority of the forecast evaluation criteria suggest that composite forecasts are superior in comparison to the individual models