Optimizing Revenue Maximization and Demand Learning in Airline Revenue Management
This work addresses the earning while learning problem for airlines, adapting an existing method to specific constraints like simultaneous price control across flights, but it is incremental as it builds on prior research.
The paper tackles the problem of balancing short-term revenue maximization with long-term demand learning in airline revenue management, showing that their adapted algorithm generates more revenue over long horizons than classical revenue-only methods.
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.