Thesis Title: Choice-based Revenue Management with Applications in Transportation

Advisors: Dr. Anton Kleywegt and Dr. He Wang

Committee members:

Dr. Alan Erera

Dr. Siva Theja Maguluri

Dr. Huseyin Topaloglu (Cornell University, ORIE)

Date and Time: 1 - 3 pm ET, Wednesday, July 8, 2020

Meeting URL: https://bluejeans.com/452306088

Meeting ID: 452 306 088 (bluejeans)

Abstract:

This dissertation consists of three studies on choice-based revenue management. All these studies are motivated by applications that arise in real-world business settings and have not been well addressed in the literature. The unifying theme of these studies is finding appropriate models to account for customer choice behavior while solving the corresponding revenue optimization problems effectively. We propose novel problem formulations, give efficient solution methods, derive insights in operational decisions, and develop near-optimal and executable strategies.

Chapter 2 considers a network revenue management problem for airlines, where airline customers tend to purchase on price: A disproportionate number of customers buy a product when it becomes the cheapest available one in the offered set of alternatives. The classic multinomial logit model cannot capture this phenomenon. We adopt a variant of the multinomial logit model and formulate a network revenue management problem for airlines. We study a deterministic approximation of the problem and show that the approximation is efficiently solvable by a small linear program. We use its solution to construct a booking limit policy and prove that the policy is asymptotically optimal.

Chapter 3 explores a dynamic load pricing problem in a truckload marketplace from a market maker’s perspective, where carriers exhibit choice behavior on loads. It is crucial for the market maker to adjust its offered price to carriers for each load based on the dynamics of supply (shipping capacity of carriers) and demand (load requests from shippers). We use a multinomial logit model to incorporate carriers’ choice behavior and formulate the market maker’s dynamic load pricing problem as a Markov decision process. We study a discrete-time fluid approximation of the problem and propose a simple pricing policy based on its solution. We show that the proposed policy is asymptotically optimal with a surprisingly small loss ratio. We also present a continuous-time fluid model and discuss the managerial insights provided by its solution.

Chapter 4 studies assortment optimization problems under a mixture of multinomial logit and independent demand models. Assortment optimization problems under mixtures of choice models are notoriously difficult. Surprisingly, we show that the single-shot assortment optimization problem under our mixture choice model is efficiently solvable. We provide a polynomial-sized linear program formulation and a combinatorial algorithm for solving such a problem. We also formulate an assortment-based network revenue management problem. We reduce a standard linear programming approximation of the problem with an exponential number of variables to an equivalent one of substantially smaller size.