Tourism forecasting by mixed-frequency machine learning

Abstract

This study aims to establish the nonlinear relationship between high frequency Baidu search volumes, low frequency tourist arrivals and dummy variables and proposes a mixed-frequency machine learning model: the Bidirectional Long Short-Term Memory Mixed Frequency Data Sampling (BiLSTM-MIDAS) model. The empirical results of forecasting weekly tourist arrivals to Kulangsu and Jiuzhaigou Valley in China demonstrate that (1) BiLSTM-MIDAS can outperform benchmark models, which is also confirmed during the COVID-19 pandemic period; (2) Compared with the MIDAS model, establishing the nonlinear relationship between high frequency Baidu search volumes, low frequency tourist arrivals and dummy variables using BiLSTM-MIDAS can improve the roles of high-frequency search engines in forecasting tourism demand. This study represents the first attempt to apply machine learning methods for tourism demand forecasting with mixed-frequency data.