Abstract
Multivariate forecasting methods are intuitively appealing since they are able to capture the inter-series dependencies, and therefore may forecast more accurately. This study proposes a multi-series structural time series method based on a novel data restacking technique as an alternative approach to seasonal tourism demand forecasting. The proposed approach is analogous to the multivariate method but only requires one variable. In this study, a quarterly tourism demand series is split into four component series, each component representing the demand in a particular quarter of each year; the component series are then restacked to build a multi-series structural time-series model. Empirical evidence from Hong Kong inbound tourism demand forecasting shows that the newly proposed approach improves the forecast accuracy, compared with traditional univariate models.