Modeling and Forecasting Daily Hotel Demand: A Comparison Based on SARIMAX, Neural Networks, and GARCH Models

Abstract

Overnight forecasting is a crucial challenge for revenue managers because of the uncertainty associated between demand and supply. However, there is limited research that focuses on predicting daily hotel demand. Hence, this paper evaluates various models’ of traditional time series forecasting performances for daily demand at multiple horizons. The models include the seasonal naïve, Holt–Winters (HW) triple exponential smoothing, an autoregressive integrated moving average (ARIMA), a seasonal autoregressive integrated moving average (SARIMAX) with exogenous variables, multilayer perceptron (MLP) artificial neural networks model (ANNs), an sGARCH, and GJR-GARCH models. The dataset of this study contains daily demand observations from a hotel in a US metropolitan city from 2015 to 2019 and a set of exogenous social and environmental features such as temperature, holidays, and hotel competitive set ranking. Experimental results indicated that under the MAPE accuracy measure: (i) the SARIMAX model with external regressors outperformed the ANN-MLP model with similar external regressors and the other models, in every one horizon except one out of seven forecast horizons; (ii) the sGARCH(4, 2) and GJR-GARCH(4, 2) shows a superior predictive accuracy at all horizons. The results performance is evaluated by conducting pairwise comparisons between the different model’s distribution of forecasts using Diebold–Mariano and Harvey–Leybourne–Newbold tests. The results are significant for revenue managers because they provide valuable insights into the exogenous variables that impact accurate daily demand forecasting.

Keywords

• daily roomnights demand prediction
• GARCH models
• GJR-GARCH
• SARIMAX
• deep learning
• ANN-MLP