Estimation Using Marginal Competitor Sales Information

ABSTRACT An abiding preoccupation for firms is understanding how customers value their products versus competitors' products. This is difficult to quantify and estimate from data as, even if competitor prices are public information, their sales are typically unobservable. However, in some industries, most prominently the hotel industry, third‐party information brokers collect and supply aggregate competitor sales information. In the hotel industry, these reports from Smith Travel Research, popularly known as STR reports, are widely subscribed to. Hotels participate by reporting their sales information and, in turn, obtain access to marginal competitor sales data, in the form of daily occupancy percentage, albeit aggregated across groups and lengths‐of‐stay. Despite its availability, this data is not widely incorporated into revenue management estimation, likely due to the lack of robust models and methodologies. In this paper, focusing mainly on the hotel industry, we develop a constrained maximum likelihood method (constrained by moment conditions) to overcome the following significant challenges in estimation of a market share model with a competitor attractiveness factor: (i) competitor data is aggregated across multiple lengths‐of‐stay with varying demand characteristics; (ii) no‐purchase data is unobservable, preventing tracking of customers who choose neither the focal firm's (we refer to as our) product nor the competitor's product; (iii) private (unobserved) group sales of competitors prior to retail sales reduce competitor capacity and influence their subsequent prices; and finally, (iv) maximizing the partial‐information likelihood function is intractable. We first evaluate our method through Monte Carlo simulations on synthetic data generated under a generalized Nash competition model. In these simulations, our method accurately recovers the true parameters to a close degree in almost all cases, exploiting the marginal competitor data. Next, we apply the method to real‐world hotel booking data and benchmark its performance against alternative approaches from the network tomography and revenue management literature.