Tourism forecasting: To combine or not to combine?

Abstract

Existing non-tourism related literature shows that forecast combination can improve forecasting accuracy. This study tests this proposition in the tourism context by examining the efficiency of combining forecasts based on three different combination methods. The data used for this study relate to tourist arrivals in Hong Kong from the top ten tourism generating countries/regions. The forecasts are derived from four different forecasting models: autoregressive integrated moving average (ARIMA) model, autoregressive distributed lag model (ADLM), error correction model (ECM) and vector autoregressive (VAR) model. All forecasts are ex post and the empirical results show that the relative performance of combination versus single model forecasts varies according to the origin–destination tourist flow under consideration, which parallels previous findings regarding the relative performance of individual forecasting methods. The results also vary with the combination techniques used. Furthermore, although the combined forecasts do not always outperform the best single model forecasts, almost all the combined forecasts are not outperformed by the worst single model forecasts. This suggests that forecast combination can considerably reduce the risk of forecasting failure. This conclusion also implies that combined forecasts are likely to be preferred to single model forecasts in many practical situations.