Applied multivariate forecasting model to tourism industry

Various forecast models can be adopted for predicting what types of tourism demand are vulnerable to wide fluctuations. This study employs the fuzzy grey model FGM(1,N) and back-propagation neural networks (BPNN) as the multivariate forecasting models. Various benchmark univariate forecasting models are also employed in this study including the naďve method, exponential smoothing model, Holt\u27s method, and linear regression. We find that the multivariate forecasting models generates more accurate forecasts than univariate models in the tourism service industry. More specifically, the GM(1,N) model was applied to choose the critical influences on tourism demand. Then, FGM(1,N) model was applied to forecast tourism demand using officially published annual data which show tourists traveling from Taiwan to the United States and to Japan during the period 1990-2003. The results showed that the FGM (1,N) outperformed the benchmark statistical methods during the out-of-sample period. Moreover, when important determinants including service price, foreign exchange rate, population, and per capita income are ranked and selected, the GM(1,N) model was improved and achieved viable performance. Finally, in terms of deciding which model to use, the general finding that can be drawn from this study appears to be that in situations involving little sampling data the grey model is superior to other traditional forecasting models. Further discussion and managerial implication can be drawn from these findings