Improving the accuracy of hotel reservations forecasting: Curves similarity and parameters processes approach

This study examines the accuracy of various time series models in forecasting daily levels of occupancy. A reservation data set of 3 hotels that include over 3450 booking curves is used to test the performance of five models in forecasting occupancy levels up to 99 days in advance. Three traditional forecasting models are compared to two new methods. The 3 benchmark models include a Stepwise Autoregression model, a High Order Polynomial model and a linear combination of their prediction. The two new methods focus on the shape of the booking curves. The first identifies the time series process of the fitted historical curves parameters, and the second applies a dissimilarity measure to identify similar curves. The results indicate that accuracy can be improved by incorporating information on the shape of past booking curves. The curve similarity model is significantly more accurate than the rest of the models that were tested.