Stochastic Peak-Load Pricing With Real-Time Demand Learning in the U.S. Airline Industry

Dynamic pricing and demand learning play an important role in airlines because (1) in low demand flights unsold tickets are of little value after departure, and (2) in high demand flights carriers may forgo important profits if the flight sells out and some relatively high willingness-to-pay consumers have to be rationed. Under a price sensitive demand, stochastic peak-load pricing suggests that at any point prior departure airlines should set higher fares in expected peak flights, where demand is more likely to exceed capacity. Moreover, in order to promote sales, lower fares should be set in expected off-peak flights. Using a unique panel of U.S. airlines fares and seat inventories, this paper shows that airlines learn about the demand as sales progress and the flight date approaches. Forecasted values of occupancy rates and sold out probabilities are employed to calibrate an ex-ante - before sales begin - distribution of demand uncertainty. Nonparametric techniques are then used to construct a latent variable to identify different expected demand states at different points prior departure. This latent variable is utilized to dictate the regime shift in a panel endogenous threshold model. Consistent with the stochastic peak-load pricing predictions, the results show that higher fares are set in the peak regime, while lower fares in the off-peak regime. The results proved to be robust to an alternative specification of a GMM dynamic panel, were the assumption of strict exogeneity is relaxed. This is the first paper to provide formal evidence of stochastic peak - load pricing in airlines or to show that airlines learn about the demand and respond to early sales