Optimizing Revenue Maximization and Demand Learning in Airline Revenue Management

Giovanni Gatti Pinheiro,Michael Defoin-Platel,Jean-Charles Regin

DOI: https://doi.org/10.48550/arXiv.2203.11065

2022-03-21

Abstract:Correctly estimating how demand respond to prices is fundamental for airlines willing to optimize their pricing policy. Under some conditions, these policies, while aiming at maximizing short term revenue, can present too little price variation which may decrease the overall quality of future demand forecasting. This problem, known as earning while learning problem, is not exclusive to airlines, and it has been investigated by academia and industry in recent years. One of the most promising methods presented in literature combines the revenue maximization and the demand model quality into one single objective function. This method has shown great success in simulation studies and real life benchmarks. Nevertheless, this work needs to be adapted to certain constraints that arise in the airline revenue management (RM), such as the need to control the prices of several active flights of a leg simultaneously. In this paper, we adjust this method to airline RM while assuming unconstrained capacity. Then, we show that our new algorithm efficiently performs price experimentation in order to generate more revenue over long horizons than classical methods that seek to maximize revenue only.

Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in airline revenue management, how to maximize short - term revenue while improving the quality of future demand behavior prediction through appropriate price experiments. Specifically, the paper focuses on the "earning while learning" (EWL) problem, that is, in the process of maximizing short - term revenue, how to collect more information about customer demand sensitivity through appropriate price changes, so as to make more accurate demand predictions in the future and ultimately maximize long - term revenue.

The paper points out that traditional revenue management systems (RMS) usually only focus on the goal of maximizing short - term revenue, ignoring the impact on long - term learning of customer demand behavior. For example, in some cases, the optimized pricing strategy may lack sufficient price changes, which makes it difficult to accurately estimate customers' price sensitivity, thereby affecting the quality of future demand predictions. This deviation may lead to a reduction in revenue. Therefore, the paper proposes a new method, aiming to optimize both revenue maximization and demand model quality simultaneously, by combining these two goals in a single objective function, with the expectation of generating more revenue in the long term.

To adapt to the airline revenue management environment, this method needs to control the simultaneous pricing strategies of multiple active flights and assumes unlimited capacity. The paper shows that their new algorithm can effectively conduct price experiments, thus generating more revenue in the long term than traditional methods that only seek to maximize revenue.