Forecast by mixed-frequency dynamic panel model

Abstract

This study presents a novel forecasting framework for panel tourism demand, utilizing a machine learning approach with mixed-frequency panel data—the first in tourism forecasting. The empirical results indicate that (a) our proposed approach, which leverages mixed-frequency panel data, significantly outperforms benchmark models in forecasting tourism demand by effectively capturing high-frequency consumer behavior information; (b) the successful capture of common information in panel data can offset the deviations brought about by individual countries' heterogeneity and improve the average accuracy of tourism demand forecasting; and (c) the machine learning approach through sparse-group least absolute shrinkage and selection operator addresses the collinearity issue in dynamic panel tourism demand forecasting and facilitates the identification of the time lag structure of influential variables.