Price Optimization in Stochastic Loss System

Abstract

Loss systems are of great importance for price optimization and revenue management even after more than a century since their first appearance. In this thesis, we analyze the optimal pricing problem for an M/M/1/1 Erlang-loss systems, and apply the model to inspect the impacts of vacancy tax regulations on short-term rental hosts. We then work on M/M/N/N loss systems while considering both advance reservation and multinomial logit (MNL) choice-model for the customers. We develop a simulation for this system and then train a machine learning (ML) model based on the outputs of this simulation to predict the utilization of each server based on different queueing parameters. Finally, we train another ML model for price optimization when the decision-maker sets the price for all servers to maximize the revenue of the whole system. We show that the presence of advance reservation decreases the utilization, consequently reducing the profit in the corresponding system.