Modeling and analysis of revenue management in airline alliances

From the most basic code-share agreements to the largest international alliances, airlines are using itineraries that bundle seats on their own flight legs with those of other airlines as a way to generate more revenue. With this growing practice comes an increasing need for researchers to look at how airlines can best manage the sharing of revenue between the airlines involved. This dissertation represents a first step toward meeting that need. In three chapters, we model and analyze the process of revenue management in airline alliances.;In Chapter 2, we begin by creating a model for the revenue management decisions faced by two partners in an alliance. We formulate a finite-horizon Markov game, over which the two airlines make accept/deny decisions on requests for itineraries on not only their own networks, but on their partners' networks, as well. The key distinctions between our model and previous single network models are the presence of gaming mechanics and the existence of both intraline itineraries --- those on a single airline's network --- and interline itineraries --- those on both partners' networks. This model is used as the basis for the remainder of the dissertation and also can serve as the framework for any future research in the area.;In the remainder of Chapter 2, we examine the game of complete information, in which both partners know each other's demand and revenue forecasts, as well as their inventory levels. Specifically, we consider three dynamic schemes that change the revenue shares received by each airline over the horizon based on the state of the system, and contrast them with existing static schemes with fixed revenue shares. We determine the equilibrium behavior for each scheme and provide insight into when each may perform well. We then show that, while well selected static schemes can perform similarly well, the dynamic schemes are far less susceptible to changes in the demand faced by the alliance.;In Chapter 3, we acknowledge the limitations of the complete information assumption due to legal, computational and competitive reasons, and focus on a game of incomplete information, in which partners do not know each other's inventory levels or forecasts. We only allow them to share bid prices, a commonly calculated quantity in modern revenue management systems. Through the selection of particular sharing scheme and a clever heuristic assumption, we are able to decouple the alliance problem into two single network problems connect by the posted bid prices. The assumption --- that the partner's bid prices will remain constant over the horizon --- is loosely based upon a result in the literature that is supported by numerical simulation. We show that the application of this heuristic leads to very favorable results despite the limited information shared between the partners. In addition, because it decouples perfectly into single network problems, any one of the numerous existing approximation schemes for that problem can be employed. This is particularly beneficial since many of the approximation methods are already being utilized in practice, facilitating the adoption of our heuristic for handling interline itineraries.;In Chapter 4, we take the constant bid price assumption one step further, looking at its application in the single airline network problem. We describe some schemes from the literature that are logically consistent with the schemes shown in Chapter 3. We also provide a new approximation method that in sample problems is shown to provide high performance (% of optimal revenue) and stable bid prices. This scheme utilizes a simultaneous-perturbation stochastic approximation method in searching for the optimal vector of bid-prices. After examining their relative performance in the single-network environment, we point out the characteristic of these schemes that may lend themselves to performing well when combined with the decoupling alliance heuristics provided in Chapter 3. Simulations of these schemes suggest that these characteristics can have greater influence on alliance revenues than the performance of the schemes in single networks.