Professor Guillermo Gallego, Crown Worldwide Professor of Engineering and the Department Head of IEDA, received the 2021 INFORMS Revenue Management and Pricing Section Prize. The awarded paper “A Markov Chain Approximation to Choice Modeling” was co-written with Professor Jose Blanchet, Professor of Management Science and Engineering at Stanford University and Professor Vineet Goyal, Associate Professor of Industrial Engineering and Operations Research at Columbia University, and was published in Operations Research (Volume 64) in 2016. This prestigious award is presented to the best contribution to the science of pricing and revenue management published in English over the past five years. This is the second time that Prof. Gallego received this distinguished award (first time was in 2005).

Congratulations!

 

Citation by INFORMS RMP Section Prize Committee: 

Discrete choice models form essential building blocks for modeling demand in revenue management and pricing applications.  The paper “A Markov Chain Approximation to Choice Modeling” proposes a new approach to discrete choice modeling, in which customer substitution behavior is directly modeled using a Markov chain (MC).  This contrasts with, yet connects to, the most common way of modeling choice behavior, which assumes that customers assign utilities to choices and then choose the one with the highest utility. Traditional utility-based approaches are designed to model customer preferences and have a long story in economics and marketing. Yet, they require making subjective assumptions about the structure of preferences and the relevant covariates that influence it. In contrast, customer substitution patterns are built upon no structural modeling assumption and are only shaped by data, becoming attractive for data-driven operational applications.  The winning nomination captures such substitution patterns directly and elegantly through Markov chains. It shows that the MC model is general, subsuming models ranging from the independent demand to the most commonly used choice models, such as the multinomial logit (MNL) model and the generalized attraction model (GAM).  The paper shows that the inferred choice probabilities are a good approximation to the true choice probabilities originated from any random utility-based choice model and derive theoretical bounds for such approximations. The paper also shows that the unconstrained assortment optimization problem is tractable under the MC model.

After its appearance, the paper has stimulated many follow-up works within the community.  A Markov chain, in addition to being intuitive, is also a mathematical object with rich history and structure.  The subsequent literature found that the MC model leads to tractable approximations of constrained assortment optimization problems (with cardinality and capacity constraints) and can also be estimated efficiently from data.  We also see ongoing work tackling price optimization under the MC model.  Because of the elegance of the model, the committee expects this work to make further inroads into practice.

The committee sees a clear stream of works that the MC model has stimulated within the community. The paper was published in 2016 and has already been extensively cited.  In sum, many approaches to modeling discrete choices have been and continue to be proposed within the literature.  But few balance tractability, mathematical rigor, and intuitive appeal quite like the MC model.