The Impact of Seasonal Unit Roots and Vector ARMA Modelling on Forecasting Monthly Tourism Flows

Abstract

The effect of imposing different numbers of unit roots on forecasting accuracy is examined using univariate ARMA models. To see whether additional information improves forecasting accuracy and increases the informative forecast horizon, the authors cross-relate the time series for inbound tourism in Sweden for different visitor categories and estimate vector ARMA models. The mean-squared forecast error for different filters indicates that models in which unit roots are imposed at all frequencies have the smallest forecast errors. The results from the vector ARMA models with all roots imposed indicate that the informative forecast horizon is greater than for the univariate models. Out-of-sample evaluations indicate, however, that the univariate modelling approach may be preferable.