Feature-Based Dynamic Pricing of Airline Ancillaries

Abstract: Airline ancillary revenue has increased substantially over the past years. Despite the increasing attention, its pricing models have mostly progressed slowly and remained simple. In this work we apply dynamic pricing models for the purpose of maximizing airline ancillary revenue. Our contributions in this thesis are threefold. Firstly, we aim to seek out high performing and robust pricing policies. Secondly, we propose a new strategy for multi-armed bandit policies called extended play. This strategy leverages on the assumption of monotonicity in willingness to pay by customers allowing the policy to play more than one arm per round. Thirdly, we propose two new algorithms, a contextual bandit policy, DEGLMUCB and a policy based on pricing under a parameterized valuation model, the OORMLP-β. They are variations of existing algorithms and are better fit to our problem setting. We evaluate our policies using historical ancillary purchase data and check for robustness with different customer behaviour settings. The main findings of this study is that the proposed contextual bandit policies with extended play were both high performing and robust to different purchase behavior settings. We also show that OORMLP-β has similar performance as the contextual bandits but can fail in some environment settings that break some of the core assumption the policy makes.