Forecasting Japanese Outbound Tourism Using Seasonal Time Series Models

Japan is one of the most important sources of international tourist arrivals to Australia and China . Using quarterly Japanese tourist arrivals to Australia and China for the period 1976(1) to 2000(2) and 1986(1) to 2000(4), respectively, this study examined the forecasting performance of various seasonal time series models, namely the Naïve Trend and Seasonal, Time Series Decomposition, Winter's Exponential, and Box-Jenkins ARIMA and SARIMA models. Before estimating the Box-Jenkins models, the augmented Dickey-Fuller tests for unit roots are used to check the stationarity of the data series. The performance of the various models are evaluated by analysing their ex post forecast accuracy using MAPE, MPE, RMSE, and Theil's U accuracy measurements. Models with the lowest forecast errors are identified as the ‘best' models. Generally, Winter's Exponential and SARIMA models have outperformed the other forecasting techniques. Goodness-of-fit measures for the ex post forecasts (produced by the ‘best' models) and the actual tourist arrivals in the same period are also estimated.