INFORMS 2021 Program Book

INFORMS Anaheim 2021

MD38

values and detours in a general setting with an arbitrary number requests. We explicitly compute the Pareto frontier for one family of city topologies, and construct it via simulation for several more networks, including one based on ridesharing data from commute hours in Manhattan. We then use these results to provide management insights and propose a two-product version of a shared rides service. MD38 CC Room 210D In Person: Fintech and OM-Finance Interface General Session Chair: Andrew Wu, Ross School of Business, University of Michigan, United States 1 - The Tokenvendor Problem: Tokenizing Cargo Reservations under Overbooking And No-shows Yunzhe Qiu, Washington University in St. Louis, St. Louis, MO, 63130, United States The container shipping industry suffers from the chronic losses caused by mismatching between the supply of liners and the demand of shippers overbooking and no-shows. We develop a model of the blockchain-based cargo reservation system, where the token is designed to be used as a booking deposit to compensate the contractual party if the other side fails to honor the booking. We propose a dynamic model for the booking deposit acceptance problem faced by the carrier with a single service slot over a fixed time horizon and characterize the optimal token reservation acceptance strategy as a downward threshold policy. Shippers with token deposits lower than a dynamic threshold should be accepted by the container liner. We propose an approximate dynamic programming algorithm, to solve the liner’s acceptance problem with computational efficiency and guaranteed performance. 2 - Measuring Utility and Speculation in Blockchain Tokens Andrew Wu, Ross School of Business, University of Michigan, Ann Arbor, MI, United States, John Silberholz Abstract not available MD39 CC Room 211A In Person: Analytical and Empirical Approaches to Healthcare Operations General Session Chair: Jayashankar M Swaminathan, University of North Carolina Chapel Hill, Chapel Hill, NC, 27599-3490, United States 1 - The Effect of Standardization On Hospital Performance Anand Bhatia, University of North Carolina Chapel Hill, UNC-CH KFBS, Chapel Hill, NC, 27599, United States, Jayashankar M. Swaminathan Healthcare services provided to patients with similar health conditions is known to vary. Evidence-based standardized protocols can be used to address such variation in care. We examine the impact of process standardization within a hospital on its performance. Our empirical analysis finds that standardization positively impacts cost of discharge, quality of outcome, and variation in outcome across departments. 2 - Self-selecting No-pay Service Delivery Strategies: A Rising Tide That Lifts All Boats? Vikrant Vaze, Dartmouth College, Hanover, NH, 03755-3560, United States, Omkar D. Palsule-Desai, Srinagesh Gavirneni, Gang Li Rapidly increasing cost structure and income inequality are major impediments in making essential services universally accessible and affordable. Service organizations are offering self-selecting no-pay options that are well-known to significantly enhance service provider profit, consumer surplus, and social welfare. Their effects on competitors have not yet been documented. We analyze a duopoly setting in which one service provider offers this strategy while the competitor does not and tabulate conditions under which these strategies increase competitor profit as well. Thus, these strategies are proving to be win-win-win- win. 3 - Oh, The Places You’ll Go! Impact Of COVID-19 Vaccination on Demand for Public Transport Huaiyang Zhong, Harvard University, Boston, MA, United States, Guihua Wang, Tinglong Dai Public transit ridership tumbled amid the COVID-19 pandemic, fueling enormous budget shortfalls and prompting slashed or eliminated services across the U.S. In this paper, we estimate the effect of COVID-19 vaccination process on the demand for public transport. Despite well-documented empirical challenges

related to causal inference of vaccination campaigns, we leverage unique features of the COVID-19 vaccine rollout to develop an instrumental variables (IV) approach. We estimate that a 1% increase in vaccination rate led to a 2.6% increase in the relative mobility in public transit centers. Our findings demonstrate the significant effect of vaccination in accelerating the recovery of public transit and provides strong support for restoring and strengthening public transit infrastructure as vaccination progresses. 4 - Collaborative Care for Diabetes And Depression Jayashankar M. Swaminathan, University of North Carolina Chapel Hill, Kenan-Flagler Business School, Operations, Chapel Hill, NC, 27599-3490, United States, Sandeep Rath Multiple clinical trials have demonstrated the benefits of a collaborative care program that integrates patients’ mental and physical healthcare within primary care. However, the financial sustainability of such collaborative care models outside trial settings has not been demonstrated. The principal challenges are: allocation of managers’ time and uncertainty over future insurance payment models. We formulate a mathematical optimization model for collaborative care for the treatment of patients with diabetes and depression towards improving clinic profits and patient QALYs. We characterize the optimal allocation of the care manager’s time. We also analyze the impact of different insurance payment models. MD40 CC Room 211B In Person: Big Data Finance and Machine Learning in Finance General Session Chair: Renyuan Xu, University of Southern California, OR Department, Los Angeles, CA ““Fx” “Chair: Zhengyuan Zhou, 1 - Presenter Tarun Chitra, Gauntlet Networks, Inc., New York, NY, United States Constant function market makers (CFMMs) such as Uniswap, Balancer, and Curve make up some of the largest decentralized exchanges on Ethereum and other blockchains. These protocols provide computational benefits for automated market makers in compute constrained environments such as blockchains. As the amount of capital and trading volume in these protocols grows past tens of billions of dollars, improving the efficiency of these systems has becomes more important. To describe what efficiency means in this context, we start with the theory of CFMMs and describe how it is closely tied to convex duality. This relationship is then related to classic payoff replications (e.g. Carr-Lee, El Karoui), which can be realized by optimized trading functions. We will describe how payoffs replicated by CFMMs can be designed to optimize capital efficiency and minimize fees. 2 - Infinitely Imbalanced Linear Classifiers with Applications To Credit Risk Mike Li, Columbia University, New York, NY, United States, Paul Glasserman In two-group linear discriminant analysis such as classifying credit risk, it is common for the data to be imbalanced: the default class is rare compared to the non-default class. Here we consider the infinitely imbalanced case where the sample size of the default class is finite and that of the non-default class grows infinitely. Under mild conditions, the regression coefficients converge to a useful limit: the exponential tilt that brings the mean of the non-default class to an exponentially weighted mean of the default class, extending a result of Owen (2007). For polynomially bounded objectives, this limit defines a distribution that is the hardest to distinguish from the non-default class; for exponential objectives with varying exponents, the linear classification based on this limit balances the trade-offs between Type 1 and Type 2 errors. 3 - Scaling Properties Of Deep Residual Networks Renyuan Xu, University of Southern California, O R. Department, Los Angeles, CA, 94720, United States, Alain-Sam Cohen, Rama Cont, Alain Rossier Residual networks (ResNets) have displayed impressive results in pattern recognition and have garnered considerable theoretical interest due to a perceived link with neural ODEs. This link relies on the convergence of network weights to a smooth function as the number of layers increases. We investigate the properties of weights trained by SGD and their scaling with network depth through detailed experiments. We observe the existence of scaling regimes markedly different from those assumed in neural ODE literature. Depending on certain features of the network architecture, we prove the existence of an alternative ODE limit, an SDE limit, or neither of these. These findings cast doubts on the validity of the neural ODE model as an adequate asymptotic description of deep ResNets and point to an alternative class of differential equations as a better description of the limit.

84