Title: Daily, weekly, monthly, and quarterly hotel room demand forecasts for Vienna across hotel classes
Authors: Ulrich Gunter - MODUL University Vienna (Austria) [presenting]
Abstract: Daily data for Vienna is employed which have been made available by the STR Share Center over the period January 1, 2010, to January 31, 2020, for the hotel classes all, luxury, upper upscale, upscale, upper midscale, midscale, and economy. The forecast variable of interest is hotel room demand (i.e., the number of rooms sold per day). As single forecast models, (1) seasonal naive, (2) ETS, (3) SARIMA, (4) Seasonal Neural Network Autoregressive (SNNAR), as well as (5) SNNAR with an additional Regressor (REG-SNNAR) are employed. As additional regressor in the REG-SNNAR model, seasonal naive forecasts of the inflation-adjusted Average Daily Rate (ADR) in euros are used. Forecast evaluation is carried out for forecast horizons $h = 1, 7, 30,$ and 90 days ahead based on rolling estimation windows (i.e., a form of time series cross-validation). As forecast combination techniques, (a) simple mean, (b) simple median, (c) least-squares weights, (d) MSE weights, and (e) MSE ranks are calculated. Based on preliminary results and a few exceptions notwithstanding, combined forecasts based on MSE (or Bates-Granger) weights and MSE ranks generally provide the highest level of forecast accuracy.