Gong Lee

Job Market Paper

Title: "Hotel Demand Estimation with and without the Assumption of Optimality  "

This paper analyzes a new database following a luxury hotel market over a 37-month period. My analysis focuses on the pricing decisions of one of these hotels, which uses a popular commercial revenue management system (RMS) to set its prices. There is a presumption that a RMS can help the hotel set optimal dynamic prices, but so far there have been no empirical studies that have been able to test this hypothesis. In order to set optimal prices, hotels must have a good knowledge of the demand for rooms in the local market in which they operate. However, there is a severe problem of endogeneity in hotel prices as a result of significant demand shocks that lead to a positive correlation between occupancy rates and average room prices in this market. There are no good instrumental variables available to deal with this endogeneity problem, and further, hotel demand and bookings is a stochastic process with customers randomly arriving to book rooms well in advance of any specific arrival date. Thus, traditional instrumental variable approaches with a linear demand curve is not appropriate or feasible for this problem. I consider two alternative approaches to estimation that can deal with the endogeneity problem and associated censoring problems due to the fact that the hotel in question only observes its own history of bookings, but not the bookings at its competitors, nor the total number of customers arriving to book rooms in this market on any particular day. Both estimation approaches use the method of simulated moments (MSM) but differ in the auxiliary assumptions used to identify the parameters of the stochastic process governing demand. One approach assumes that the hotel sets its prices optimally as a best response to the prices of its competitors. The other approach relaxes this optimality assumption and estimates demand with a two-step semi-parametric approach. These two approaches produce different parameters, resulting in alternative price policies, which I use to compare the estimated price elasticities and other demand parameters for different types of consumers. Using the estimated demand process from the two-step semiparametric MSM estimator that relaxes the assumption of optimal pricing, I solve the dynamic programming problem (DP) to determine the optimal pricing policy implied by these estimates. I find that the optimal pricing strategy from the DP solution differs significantly from the pricing strategy the hotel actually uses. For non-busy days, the optimal predicted pricing strategy starts with significantly higher prices than what the hotel actually charges from 45 to 20 days in advance of arrival, but then drops and falls significantly below what the hotel charges in the last 20 days prior to arrival. The model predicts this counter-factual optimal pricing strategy raises revenues and occupancy by 10% and 9%, respectively, over what the hotel actually achieves on non-busy days.