LAUSR

Abstract The forecasting literature shows that when a number of different forecasters produce forecasts of the same variable it is almost always possible to produce a better forecast by linearly combining the individual forecasts. Moreover, it is often argued that a simple average of the forecasts will outperform more complex combination methods. This paper shows that, analytically, nonlinear combinations of forecasts are superior to linear combinations. Empirical results, based on comparisons of real GDP growth projections with outturns for the euro area using time-varying-coefficient estimation, confirm that analytical result, especially for periods marked by structural changes.