A note on forecasting euro area inflation: leave- $$h$$ h -out cross validation combination as an alternative to model selection

This paper evaluates the jackknife (leave- $$h$$ h -out) technique of averaging individual forecasting models. It offers an initial empirical attempt to apply the leave- $$h$$ h -out cross-validation (CV) combination scheme to time series forecasting. The problem of obtaining weights for combining individual autoregressive forecasting models is solved by minimizing the CV criterion (as opposed to the simple (equally weighted) average of the considered benchmark models). The procedure is illustrated on the euro area yearly inflation rate. The recursive pseudo-out-of-sample forecasts are estimated with the forecasting period starting at the beginning of the recent economic crisis (August 2008). The results indicate CV’s dominance with respect to benchmark models for longer forecast horizons (more than 4 months ahead). Moreover, a sensitivity analysis is performed in order to examine whether the obtained results are influenced by the choice of forecast horizon or the out-of-sample cut-off point. In the vast majority of cases, the CV combination superiority is indifferent to the forecast horizon selection. Its dominance is also mostly not influenced by the cut-off point choice. Copyright Springer-Verlag Berlin Heidelberg 2015