Forecasting international city tourism demand for Paris: accuracy of uni- and multivariate models employing monthly data

The purpose of this study is to compare the predictive accuracy of various uni- and multivariate models in forecasting international city tourism demand for Paris from its five most important foreign source markets (Germany, Italy, Japan, UK and US). In order to achieve this, seven different forecast models are applied: EC-ADLM, classical and Bayesian VAR, TVP, ARMA, and ETS, as well as the naïve-1 model serving as a benchmark. The accuracy of the forecast models is evaluated in terms of the RMSE and the MAE. The results indicate that for the US and UK source markets, univariate models of ARMA(1,1) and ETS are more accurate, but that multivariate models are better predictors for the German and Italian source markets, in particular (Bayesian) VAR. For the Japanese source market, the results vary according to the forecast horizon. Overall, the naïve-1 benchmark is significantly outperformed across nearly all source markets and forecast horizons.