Network revenue management with cancellations and no-shows

摘要

Airline customers often cancel bookings or may not show up for their flights. To overcome this problem, airlines accept overbookings to enhance their revenues. Even though many studies have been conducted on this practice, problems related to overbooking persist. The number of passengers who are forced to give up their seats increase every year at a high rate, making overbooking costly and as well as increasing customer dissatisfaction. There are only a few studies related to cancellations and no-shows linked with the issue of airline revenue management. The industry relies on static overbooking models and related heuristics. These models give the maximum number of overbookings based on the current rate of cancellations. Some empirical studies have shown that dynamic models can provide increased revenues compared to static models. Some studies adopt the framework of dynamic seat inventory control with and without multiple fare classes. The dynamic problem that is class-independent is tractable. However, the assumption of class independence is not realistic. The Markov chain decision process model should take into account the number of seats reserved in each class. The authors propose two types of demand models to manage revenue management problems with cancellations and no-shows. One model is called the independent-demand model in which booking request rates are exogenous. The other model is called the choice-based-demand model in which booking request rates for different products are endogenous depending on the product type and customer behavior. The customers arrive in both models according to a non-homogenous Poisson process and each customer may cancel before the terminal time, become a no-show, or take the booked flight. The two resulting models are not tractable, so a corresponding deterministic model is used in place of both models that is called the fluid model. Based on the solution for the fluid model, the solution to the stochastic (discrete) model is derived. The fluid model can be used to construct a booking policy for the original stochastic discrete model.