Forecasting tourist arrivals using multivariate singular spectrum analysis

Abstract

The accurate forecasting of tourist arrivals has become a necessity for destination managers and tourism businesses. Singular spectrum analysis (SSA) has been applied in other areas, although its application in tourism demand is limited to SSA using a single univariate time series. New developments in the field extend the univariate framework into a multivariate SSA (MSSA). This article aims to forecast tourist arrivals from five continents to South Africa using MSSA and to compare the forecasting accuracy with that of univariate SSA as well as the baseline seasonal naïve model. The results show that in all but one case, MSSA leads to improved forecasting accuracy compared to univariate SSA and that these improvements are especially prevalent when forecasting over longer time horizons.