Combining linear and nonlinear model in forecasting tourism demand

Much research shows that combining forecasts improves accuracy relative to individual forecasts. However, existing non-tourism related literature shows that combined forecasts from a linear and a nonlinear model can improve forecasting accuracy. This paper combined the linear and nonlinear statistical models to forecast time series with possibly nonlinear characteristics. Real time series data sets of Taiwanese outbound tourism demand were used to examine the forecasting accuracy of the combination models. The forecasting performance was compared among three individual models and six combination models, respectively. Among these models, the normalized mean square error (NMSE) and the mean absolute percentage error (MAPE) of the combination models were the lowest. The combination models were also able to forecast certain significant turning points of the test time series. Thus, this paper suggests that forecast combination can achieve considerably better predictive performances and show promising results in directional change detects ability in the tourism context. Besides, the empirical results also clearly show that how a high forecasting accuracy and an excellent directional change detect ability could be achieved by the SVR combination models.