Fluid arrivals simulation for choice network revenue management

Since the beginning of revenue management, simulation has been used to estimate the expected revenue resulting from an availability policy. It has also been used to verify the quality of forecasts by projecting them onto past availability policies. Recently, it has been used in simulation-based optimization approaches to find the best policy. Simulation thus has a central role in revenue management. We focus on the choice network revenue management (CNRM) problem that incorporates multiple resources and customer behavior. The traditional CNRM simulation is based on discrete customer arrivals; we propose a new approach based on fluid arrivals. Our estimator is biased, but we observe that the bias is often insignificant in practice and appears to be asymptotically null. Our approach consistently outperforms the traditional simulation in terms of estimation time and is thus a better choice for large instances. We also prove that it is equivalent to an approximation for the CNRM availability policy optimization problem. This equivalence limits the value of simulation-based optimization methods but allows us to propose heuristics to rapidly support the optimization.