Seasonal Behaviour of Monthly International Tourist Flows: Specification and Implications for Forecasting Models

Recent studies have found that seasonality in international tourist arrivals is more likely to be stochastic than deterministic. The purpose of this paper is to examine the effect of different specifications of seasonality on forecasting performance. The authors compare forecasts generated from the regression-based model, seasonal ARIMA model, and Harvey's structural time series model, using a number of monthly international tourist arrivals to Australia. They find seasonality to be deterministic using the HEGY test for seasonal unit roots for all cases, while the Caner test finds seasonality to be stochastic for almost all cases. The use of descriptive measures suggests that stochastic seasonality is more appropriate. The authors find evidence that stochastic treatment of seasonality does not improve forecasting performance, regardless of the presence of seasonal unit roots. The regression-based model tends to generate superior forecasts when seasonality is treated as deterministic.