A re-evaluation of the combination and adjustment of forecasts

Abstract

The forecasting literature suggests that forecasts based on different information sets may be combined to produce predictions with greater accuracy than that of the individual forecasts. In addition, it has been suggested that correction for bias and regression components of forecast errors can also improve the accuracy of a forecast by the mean squared error criterion. In this paper two forecasts (univariate time-series and survey forecasts) using vastly different information sets and methodologies are evaluated and compared, and their performance singly, in combination and corrected for bias and regression components of forecast errors is examined. It is found that combined or adjusted forecasts do not necessarily perform better than individual uncorrected forecasts in out-of-sample evaluations.