Bootstrapping Australian inbound tourism

Abstract

This paper adopts a model-free approach to forecasting monthly international tourist arrivals to Australia from four major origin countries: New Zealand, UK, the USA, and China. While most researchers use parametric methodologies to model tourism demand, this study proposes a non-parametric approach that employs a 'Partitive Simulation Process' or PASIP by partitioning the original monthly time series into 12 sub-series according to the month. Both ordinary and time-weighted non-parametric bootstraps are used, to resample the observed samples 2,000 times, to estimate the underlying population statistics. We then conduct PASIP's forecasts and compare them with ARIMA forecasts. There are four significant observations. First, the partitive process is a viable way of handling seasonality. Second, PASIP performs well in predicting the turning points and data trends. Third, weighted PASIP generally outperforms non-weighted PASIP in terms of forecast errors. Fourth, PASIP produces smaller forecast errors than ARIMA for UK, USA, and China data.