Kevin Williams, Yale University
ABSTRACT: We investigate the ability to learn as well as the consequences of demand learning in the airline industry using rich data provided by a large international air carrier based in the United States. The ability to accurately predict future demands has important welfare consequences when capacity is scarce and perishable -- for example, if a firm incorrectly believes that an off-peak demand flight is peak, prices will be too high and output inefficiently low. Using detailed forecasting information, we document an upward bias in the airline's forecasting methodology -- a bias that decreases as the departure date approaches. To quantify the welfare effects of the airline's revenue management system, we propose a Poisson-random coefficients logit demand model and Bayesian estimation approach that leverages granular search and sales data. The method accommodates sparse sales and endogenous prices. With the model estimates, we compare market outcomes under recovered firm beliefs about demand with and a model of demand learning through consumer search activity.