Resumen
The purpose of this study is to emphasize the advantages of Mahalanobis distance in assessing the overall accuracy of inflation predictions in Romania when two scenarios are proposed at different times by several experts in forecasting or forecasters using data from a survey (F1, F2, F3 and F4). Mahalanobis distance evaluates accuracy by including at the same time both scenarios and it solves the problem of contradictory results given by different accuracy measures and by separate assessments of different scenarios. The own econometric model was proposed to make inflation rate forecasts for Romania, using as explanatory variables for index of consumer prices the gross domestic product, index of prices in the previous period and the inverse of unemployment rate. According to Mahalanobis distance, in 2012 and 2013 F1 registered the highest forecasts accuracy distance. The average distance shows that F1 predicted the best the inflation rate for the entire period. F2 provided the less accurate prognoses during 2011-2013. According the traditional approach, based on accuracy indicators that were evaluated separately for the two scenarios, F1 forecasts provided the lowest mean absolute error and the lowest root mean square error for both versions of inflation predictions. All the forecasts of the inflation rate are superior as accuracy of naïve predictions. However, according to U1 Theil?s coefficient and mean error, F3 outperformed all the other experts and also the forecasts based on own econometric model.