Many markets are characterized by fixed and perishable capacity, a gradual arrival of consumers, and uncertainty concerning the overall magnitude of demand. Some examples include air travel, hotels, trains, and entertainment or sports events. In order to manage these situations, many firms undertake dynamic pricing, often called “revenue management” by industry practitioners. Prices may adjust based on changes in the elasticity of demand or changes in the opportunity cost of remaining capacity. The latter may occur due to stochastic demand, but also because firms may acquire information over time. For example, a flight may be a peak demand flight and this is not initially known. Empirical studies of dynamic pricing with demand learning are limited, however, due to the difficulty in modeling the complex dynamic problem facing firms and the unavailability of detailed data on prices, quantities, and firm beliefs. This project conducts an empirical investigation of revenue management (RM) and quantifies the welfare impacts of dynamic pricing with demand learning in the airline industry.
In our first project, we investigate the ability to learn about demand over time and quantify the consequences of demand learning on welfare. For example, if a firm incorrectly believes that an off-peak flight is peak, prices will be too high and output inefficiently low. On the other hand, demand learning may allow a firm the ability reserve capacity for those with the highest willingness to pay and extract their surplus through targeted prices. We propose a new methodology to estimate demand in markets with sparse sales and endogenous prices by combining features of stochastic demand models in operations research with random coefficients demand models that are commonly used in industrial organization and quantitative marketing. Although we tailor our specifications to the airline industry, our modeling approach can be applied to any setting where researchers have access to search and sales information. E-commerce is a relevant application. With demand estimates, we compare how the firm allocates capacity in practice to the outcomes of a Bayesian learning model in which the firm updates its beliefs about future demands by observing consumer arrivals. We contrast these findings to a model in which the firm does not update its beliefs over time.
Requisite Skills and Qualifications:
- Justin Ye
- Alice Geng