Forecasting Tourism Using Univariate and Multivariate Structural Time Series Models

Abstract

Tourism demand forecasting remains an important research area, as the search for more accurate forecasting methods continues. In particular, there is concern that many methods do not improve upon a simple naïve process. Structural time series models have shown significant potential as both univariate and explanatory forecasting tools. Inbound tourism to New Zealand from Australia, Japan, the UK and the USA disaggregated by purpose of visit is analysed, using both univariate and multivariate structural time series models, and their respective forecasting accuracy is compared. The naïve ‘no change’ model is used for benchmark comparison purposes. The structural time series model outperforms the naïve process, but the causal structural time series model does not generate more accurate forecasts than the univariate model.