Learning and optimizing through dynamic pricing
- 172 Downloads
Many airlines have been actively looking into class-free inventory control approaches, in which the control policy consists of dynamically varying prices over a continuous interval rather than opening and closing fare classes. As evidenced both in literature and in practice, one of the big challenges in this setting is the trade-off between policies that learn the demand parameters quickly and those that maximize expected revenue. Starting in a typical single-leg airline revenue management context, we investigate the applicability of recent advances in the area of optimal control with learning. We consider a demand model where customers’ maximum willingness-to-pay has a Gaussian distribution and we analyze several estimation and pricing approaches that include the expectation–maximization and a scheme of active generation of price variability. We show that our model ensures discovery of the underlying customer behavior while providing an appropriate level of expected revenue via a simulated example.
KeywordsDynamic pricing Exploration–exploitation Revenue management Regret Expectation–maximization Simulation
We are grateful to Dr. Darius Walczak and the two anonymous referees for their valuable comments that helped improve this paper.
- Dempster, A.P., N.M. Laird, and D.B. Rubin. 1977. Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society Series B 39 (1): 1–38.Google Scholar