INFORMS 2021 Program Book

INFORMS Anaheim 2021

MB04

Monday, 7:45AM-9:15AM

MB03 CC Ballroom C / Virtual Theater 3 Hybrid Markov Lecture Sponsored: Applied Probability Society Sponsored Session Chair: Rhonda L. Righter, University of California-Berkeley, Berkeley, CA, 94720-1777, United States 1 - Fragmenting Financial Markets Darrell Duffie, Stanford University, Graduate School of Business, 518 Memorial Way, Stanford, CA, 94305-5015, United States This talk on financial market design addresses the costs (and sometimes the benefits) of fragmenting trade across multiple venues.Size discovery trading crosses buy and sell orders, with no bid-ask spread and no price impact, by exploiting the price determined on a separate exchange market.Although popular in practice, size discovery reduces the depth of exchange markets and, as modeled, worsens overall allocative efficiency.On the other hand, fragmenting trade in the same asset across multiple exchanges can improve allocative efficiency.This talk draws from research with Samuel Antill, Daniel Chen, and Haoxiang Zhu. 2 - Discussant Mathieu Rosenbaum, Ecole Polytechnique, France MB04 CC Ballroom D /Virtual Theater 4 Hybrid Predictive Analytics Applications Sponsored: Artificial Intelligence Sponsored Session Chair: Yixin Lu, The George Washington University, Washington, DC, 20052, United States Chair: Francesco Balocco, Rotterdam School of Management Erasmus University, Rotterdam, 3011 ZX, Netherlands 1 - Tech Tax: Ad Exchanges’ Fees in Display Advertising Francesco Balocco, Rotterdam School of Management Erasmus University, Rotterdam, Netherlands, Yixin Lu, Ting Li We study the Ad Exchanges’ (ADX) fee optimization problem under the two dominant mechanisms in the display advertising market: the waterfall and the header bidding mechanism. We address two research questions: (1) What are the welfare implications of ADXs’ fee structures? (2) How do ADXs’ optimal fee structures evolve under different market configurations? Our study contributes to both theory and practice of digital advertising. First, to the best of our knowledge, this study is among the first to examine the welfare implications of ADXs’ fee structures under different market mechanisms. Second, our findings shed light on the underlying drivers for the publisher’s move from the waterfall mechanism to the header bidding mechanism. Finally, our model allows ADXs to perform policy counterfactuals, providing useful implications for their decisions on fee structures. 2 - Investigating the Willingness to Pay For Enhanced Mobile Internet Services: Evidence From a Mobile Network Upgrade Yi Zhu, University of Minnesota at Twin Cities, Minneapolis, MN, United States, Jason Chan, Xuan Bi We investigate consumers’ willingness to pay (WTP) for two mobile internet pricing models, the speed-based model and the data-consumption-based model. We examine consumer sentiment toward unexpected slow network speed and data overuse problems, which provides valuable information for understanding customers’ WTP for increasing network speed or data allowance. Our empirical strategy leverages the staggered introduction of 4G network across various districts in a metropolitan Asian city and a quasi-experimental setup. We find that consumers are more willing to pay for increasing network speed than data allowance in the new mobile internet era. We also show that substantial heterogeneity in consumer sentiment is explained by consumers’ income and age. These insights can inform relevant stakeholders of optimal responses around future mobile network pricing models.

MB01 CC Ballroom A / Virtual Theater 1 Hybrid AAS Special Speaker Talk Sponsored: Aviation Applications Sponsored Session Chair: Alexandre Jacquillat, MIT Sloan School of Management, Cambridge, MA, 2142, United States 1 - Introduction of AI/ML Capabilities into Airline Industry Sergey Shebalov, Sabre Holdings, Southlake, TX, 76092, United States AI/ML capabilities are being widely adopted across many industries. We will share practical experience of introducing these capabilities into airline industry. There are several key properties of AI/ML that generate incremental value compare to decision support approaches used in the past. We describe specific use cases to illustrate these properties and discuss typical challenges and roadblocks for successful implementation of AI/ML capabilities. We will also look beyond creating ML models to into data management and MLOps areas that are crucial for AI/ML adoption. Finally, we’ll make a few suggestions on the role of academic community in this process and describe a path toward closer collaboration between academia and industry. MB02 CC Ballroom B / Virtual Theater 2 Hybrid Inverse Optimization Sponsored: OPT/Optimization Under Uncertainty Sponsored Session Chair: Taewoo Lee, University of Houston, Houston, TX, 77204-4008, United States 1 - Learning Personalized Diabetic Retinopathy Screening Preferences Fariha Kabir Torsha, University of Houston, Houston, TX, United States, Taewoo Lee Diabetic retinopathy (DR) is the leading cause of vision loss in working-age Americans. Due to the asymptomatic early stages of DR, the American Diabetes Association recommends annual eye screening exams for all diabetic patients. However, not all patients are screened annually; compliance rate varies significantly across different types of patients, typically within the range of 20- 60%. In this study, we model the patient’s screening decision-making process as a Markov decision process (MDP) and use inverse optimization to infer the patient’s reward function from his/her past screening decisions. We then use the inferred reward function to generate personalized screening decisions. 2 - A Penalty Block Coordinate Descent Algorithm for Data-driven Inverse Convex Optimization Rishabh Gupta, University of Minnesota, Minneapolis, MN, United States, Qi Zhang We consider inverse convex optimization where the goal is to jointly infer the unknown objective and constraint parameters of a convex NLP from noisy observations. We formulate the problem as a bilevel program and apply a KKT- based approach to obtain a single-level reformulation. The resulting nonconvex nonlinear problem is solved with an approximate block coordinate descent method. We show that the proposed algorithm is guaranteed to converge to a stationary point for several important classes of forward problems such as convex QCQPs and geometric programs. Numerical experiments on synthetic datasets demonstrate the computational advantage of our method against standard commercial solvers. 3 - Optimality-Based Clustering Taewoo Lee, University of Houston, Houston, TX, 77204-4008, United States, Zahed Shahmoradi Clustering is a well-known technique to group a set of data points into smaller clusters such that the data points in the same cluster are closer to each other than to those in other clusters based on some similarity function. We propose a new clustering approach, called optimality-based clustering, that clusters data points based on their encoded decision preferences. We model the problem as a mixed- integer program and propose efficient heuristics.

45