INFORMS 2021 Program Book

INFORMS Anaheim 2021

TC29

TC29 CC Room 207C

TC31 CC Room 208A In Person: Mechanism Design, Networks, and New Markets General Session Chair: Francisco Castro, UCLA Anderson School of Management, Los Angeles, CA, 90024, United States 1 - Dynamic Moral Hazard with Adverse Selection: A Pontryagin Approach Feifan Zhang We study the optimal incentive scheme for a long-term project with both moral hazard and adverse selection. The moral hazard issue is due to the fact that the agent’s effort, which increases the arrival rate of a Poisson process, is not observable by the principal. In addition, the agent’s effort cost, which needs to be reimbursed by the principal, is also the agent’s private information. This gives rise to the adverse selection problem. The principal needs to design the optimal menu of contracts, each of which is chosen by the agent with a specific effort cost. We fully characterize the optimal menu in the case of two types of agents. Specifically, the agent with a lower cost is offered a probation contract, which confirms the agent’s type if there is an arrival during a probation period; the agent with the higher cost is offered a sign-on-bonus contract with an immediate direct initial payment. We then explore the more general case with continuous types of agents. In particular, we provide an easy-to-compute upper bound on the principal’s utility. The upper bound computation also yields a feasible menu of probation and sign-on-bonus contracts, and the corresponding lower bound it generates. We further provide a condition which can be used to verify whether the upper and lower bounds coincide, implying the optimality of our feasible menu of contracts. Numerical studies confirm that the verficiation condition almost always holds for commonly used probability distributions of the effort cost. 2 - Mechanism Design under Approximate Incentive Compatibility Francisco Castro, UCLA Anderson School of Management, 1315 Devon Avenue, Los Angeles, CA, 90024, United States We extend the classical Myerson setting to the case where the buyer is not a perfect optimizer and only -incentive compatibility is required. Our paper is the first to study the design of optimal mechanisms in the space of approximate IC mechanisms and to explore how much revenue can be garnered by moving from exact to approximate incentive constraints. We show that deterministic mechanisms are not optimal and that randomization is needed. We then establish that no mechanism can garner gains higher than order 2/3. This improves upon known results that imply maximum gains of 1/2. Furthermore, we construct a mechanism that is guaranteed to achieve order 2/3 additional revenues, leading to a tight characterization of the revenue implications of approximate IC constraints. Importantly, to find the optimal mechanism it is necessary to optimize over best reporting functions. 3 - Providing Data Samples for Free Kimon Drakopoulos, University of Southern California, Los Angeles, CA, 90305-1028, United States We consider the problem of a data provider (Seller of information) who sells information to a firm (Buyer of information) regarding an unknown (to both parties) state of the world. Traditionally, the literature explores one-round strategies for selling information due to the Seller’s hold-up problem: once a portion of the dataset is released, the Buyer’s estimate improves and as a result the value of the remaining dataset drops. In this paper we show that this intuition is true when the Buyer’s objective is to improve the precision of her estimate. On the other hand, we establish that when the Buyer’s objective is to improve operational decisions (e.g. better pricing decisions on a market with unknown elasticity) and when the Buyer’s initial estimate is misspecified, one-round strategies are outperformed by free-sample selling strategies and dynamic pricing. TC32 CC Room 208B In Person: Algorithmic Competition and Collusion in Revenue Management General Session Chair: Chamsi Hssaine, Cornell University, Ithaca, NY, 14850, United States 1 - Pseudo-Competitive Games Chamsi Hssaine, Cornell University, Los Angeles, CA, 90025-5692, United States, Vijay Kamble, Siddhartha Banerjee We study algorithmic price competition in a duopoly under a model of satisficing customer behavior. In this model, customers consider firms in some exogenously determined order of preference until they find a price that satisfies an ideal surplus target, choosing the lowest price they can afford if every firm fails to satisfice. We exhaustively characterize the equilibrium landscape of the game, and show that it is frequently plagued by strictly-local Nash equilibria, an outcome

In Person: Practice Curated: Analytics Enabling Healthcare Access through Telehealth/Enterprise Revenue Management at 84.51 General Session Chair: Yifan Liu, 84.51°, Cincinnati, OH 1 - Service System Design of Video Conferencing Visits with Nurse Assistance Xiang Zhong, University of Florida, Gainesville, FL, 32612-1701, United States, Xiaojie Wang, Yongpei Guan Video-conferencing (VC) clinical visits (aka telehealth service) are gaining popularity over the past decades. We explore the implementability of VC visits with nursing services using a game-theoretic model, and investigate the impact of different pricing schemes (discriminative pricing based on patient characteristics vs. non-discriminative) on patients’ care choices between VC visits and in-person visits. Our results identify the conditions where the interest of the social planner and the medical institution could be aligned, and highlight that, compared to a uniform price of VC visits which seems fair, a discriminative pricing strategy can be more beneficial for patients and the medical institution alike. The insurance coverage of telehealth-related services is important to promote the adoption of telehealth by patients and care providers. 2 - Regular Price Decision Making under Uncertainties Yifan Liu, 84.51°, Cincinnati, OH, United States In retail businesses, a decision maker often needs to make long-term strategic pricing decisions (regular price) without finalizing a complete set of short-term tactical decisions such as promotional price, ads plan and display plan. We present a framework using a stochastic programming model to help Kroger make this type of regular price decision that hedges against future uncertainties while maintaining logic on shelf and improving financial metrics. TC30 CC Room 207D In Person: Modeling and Analytics for Heterogeneity in System Informatics Joint Session Chair: Minhee Kim, University of Wisconsin-Madison, Madison, WI, 53706-1539, United States 1 - Multi-modal Predictive Model for Persistent Post-traumatic Headache Nathan B. Gaw, Georgia Institute of Technology, Atlanta, GA, 85258-2222, United States, Catherine Chong, Todd Schwedt, Visar Berisha, Teresa Wu, Katherine Ross, Gina Dumkrieger, Jianwei Zhang, Simona Nikolova, Jing Li Each year, there are approximately 2 million individuals diagnosed with mild traumatic brain injury (mTBI). Post traumatic headache (PTH) is the most common symptom following mTBI, which can either resolve or continue into persistent PTH (PPTH). There is a need to determine biomarkers that can be used to predict resolution or persistence of PTH in order to allow for timely treatment of the condition. The current study builds a multi-modality predictive model that combines clinical measures, medical images, and longitudinal speech data to predict at-risk patients and identify highly relevant biomarkers. 2 - Covariate Dependent Sparse Functional Data Analysis Minhee Kim, University of Wisconsin–Madison, Madison, WI, 53705, United States, Kaibo Liu We propose a method to incorporate intrinsic covariate information into sparse functional data analysis. The method aims at cases where each unit has a small number of longitudinal measurements and records intrinsic covariates. Unlike external covariates which may change over time, intrinsic covariates represent the basic nature of a unit and static. Existing methods often do not use covariate information or only model the additive effects of dynamic external covariates. Yet, the incorporation of intrinsic covariate information into functional data analysis can significantly improve modeling and prediction performance. The proposed method decomposes the variation of measurements into the variation coming from intrinsic covariates and the variation left conditioned on intrinsic covariates. We also develop a bootstrapping covariate selection algorithm.

120