Title: Exploring the predictive ability of LIKES of posts on the Facebook pages of four major city DMOs in Austria
Authors: Ulrich Gunter - MODUL University Vienna (Austria) [presenting]
Irem Onder - MODUL University Vienna (Austria)
Stefan Gindl - MODUL University Vienna (Austria)
Abstract: Using data for the period 2010M06 - 2017M02, the aim is to investigate the possibility of predicting total tourist arrivals to four Austrian cities (Graz, Innsbruck, Salzburg, and Vienna) from LIKES of posts on these destinations DMO Facebook pages. Google Trends data are also incorporated in investigating whether forecast models with LIKES and/or with Google Trends deliver more accurate forecasts. To capture the dynamics in the data, the ADL model class is employed. Taking into account the daily frequency of the original LIKES, the MIDAS model class is also employed. While time-series benchmarks from the naive, ETS, and ARMA model classes perform best for Graz and Innsbruck across horizons and accuracy measures, ADL models incorporating only LIKES or both LIKES and Google Trends generally outperform their competitors for Salzburg. For Vienna, the MIDAS model including both LIKES and Google Trends produces the smallest RMSE, MAE, and MAPE values for most horizons. Therefore, for at least two of the four Austrian cities under scrutiny, incorporating complementary information originating from two different web-based predictors is worthwhile in order to produce more accurate tourism demand forecasts. In addition, forecast encompassing tests relative to the naive-1 benchmark reject their null hypothesis at least at the 10\% significance level in 18 out of 24 cases across cities and horizons, thus making the use of more sophisticated forecast models meaningful in the first place.