Abstract
The last three decades have seen the booming of the research on combination forecasts in various economics and business fields. However, the applications of the methodologies in the tourism context are still rare. This study aims to improve the accuracy of tourism demand forecasting by combining individual forecasts generated from modern econometric models and time series models. An empirical study which focuses on UK outbound leisure tourism demand for seven major destinations is carried out. The combinations are based on the competing forecasts generated from seven individual forecasting techniques: seasonal Naive model, SARIMA model, reduced ADLM, Wickens-Breusch ECM, Johansen ECM, VAR and TVP models. The combination methods examined in this study are the simple average combination method, the variance-covariance combination method, the discounted mean square forecast error (MSFE) method, the Granger-Ramanathan Regression method, the Shrinkage method and the time varying parameter (TVP) forecast combination method. This study has also provided a systematic comparison of seasonal forecast accuracy among advanced econometric techniques, along with the commonly used time series benchmarks. The empirical results suggest that overall combination forecasts play an important role in the improvement of forecasting accuracy in that they are superior to the best of the individual forecasts. The discounted MSFE method turns out to be the best combination method among six, and the length of forecasting horizons does not appear to have an effect on the performance of combination forecasts. The empirical evidence also shows that more sophisticated combination forecasts which take the historical performance of the individual forecasts into consideration perform better than simple average combination method. For the shrinkage forecasts, the greater the shrinkage, the greater the accuracy of the resulting combination forecast. The TVP combination method, which permits the greatest time variation in weights, exhibits the poorest performance in this study. And encompassing test is found to contribute to the improved accuracy of combination forecasts, therefore it is necessary to conduct encompassing test before individual forecasts are combined. Another finding is that different treatment of seasonality affects forecasting performance of alternative models and the seasonal unit root pre-test can improve forecast accuracy. The empirical study provides further evidence for the efficiency of combination forecasts in that combination forecasts should be preferred over single model forecasts in future tourism forecasting practices.