An empirical study of forecast combination in tourism

Abstract

The performance of forecast combination techniques is explored at different time horizons in the context of tourism demand forecasting. Statistical comparisons between the combination and single-model forecasts show that the combined forecasts are significantly more accurate than the average single-model forecasts across all forecasting horizons and for all combination methods. This provides a strong recommendation for forecast combination in tourism. In addition, the empirical results indicate that forecast accuracy does not improve as the number of models included in the combination forecasts increases. It also appears that combining forecasts may be more beneficial for longer-term forecasting.