INFORMS 2021 Program Book

INFORMS Anaheim 2021

WD31

WD24 CC Room 205A In Person: Supply Chain Data Analytics Applications General Session Chair: Ahmet Colak, Clemson University, Pendleton, SC, 29670, United States 1 - Capacity Planning and Online Order Fulfillment with Integrated Logistics Pin-Yi Chen, Massachusetts Institute of Technology, Cambridge, MA, 02141-1904, United States, Tolga Cezik, Daniel Chongli Chen, Tamar Cohen-Hillel, Stephen C. Graves The order fulfillment decision for an online retailer includes determining the fulfillment center from which to source an order as well as the transportation route for shipping the order to the customer. Increasingly, online retailers are relying on their own integrated logistics operations for their shipping, in addition to third-party shipping. For this case we propose a framework for capacity planning of the integrated logistics operations in conjunction with dynamic order fulfillment decisions. A realistic capacity plan of a large-scale network is realized with our framework. 2 - An Empirical Analysis of Feature-based Pricing in the Automobile Industry Choi Hojun, Northwestern University, Evanston, IL, United States, Achal Bassamboo, Ahmet Colak, Sina Golara Auto consumers pay close attention to a car’s pricing, functionality, and attributes, and dealerships consider this aspect of consumer behavior when determining their prices and which features to emphasize. We explore whether listing similar features improves dealerships’ performance based on 10,000 auto dealerships from a national sample. Our cars.com dataset of 2 million cars is obtained daily from August until December 2020. W e estimate the main effects by regressing price and listing duration on the car, brand, dealership, location, and feature characteristics. Our findings can provide more concrete insights into customers’ behaviors regarding redundant information of an item. WD27 CC Room 206B In Person: Faster Conditional Gradient Methods General Session Chair: Alejandro Carderera, United States 1 - Universal Conditional Gradient Sliding for Convex Optimization Trevor Squires, Clemson, Clemson, SC, 29678-1719, United States, Yuyuan Ouyang We present a first-order projection-free method, namely, the universal conditional gradient sliding (UCGS) method, for solving -approximate solutions to convex differentiable optimization problems. For objective functions with Hölder continuous gradients, we show that UCGS is able to terminate with -solutions with at most O((1/ )2/(1+3v)) gradient evaluations and O((1/ )4/(1+3v)) linear objective optimizations, where v is the exponent and of the Hölder condition. In the weakly smooth case when v is in (0,1), both complexity results improve the current state-of-the-art O((1/ )1/v) result achieved by the conditional gradient method. In the smooth case when =1, UCGS matches the state-of-the-art complexity result and adds more features allowing for practical implementation. Furthermore, UCGS runs without explicit knowledge of the smoothness information. 2 - The Primal-Dual Davis-Yin Algorithm(s) Adil Salim, King Abdullah University of Science and Technology (KAUST), King Abdullah University of Science and Technology, Thuwal, Saudi Arabia We introduce new primal-dual algorithms to minimize the sum of three convex functions, each having its own oracle. Namely, the first one is differentiable, smooth and possibly stochastic, the second is proximable, and the last one is a composition of a proximable function with a linear map. By leveraging variance reduction, we prove convergence to an exact solution with sublinear or linear rates, depending on strong convexity properties. The proposed theory is simple and unified by the umbrella of stochastic Davis-Yin splitting, which we first design in this work. Our theory covers several settings that are not tackled by any existing algorithm.Joint work with Laurent Condat, Konstantin Mishchenko and Peter Richtárik.

WD30 CC Room 207D In Person: Recent Developments on Modeling Financial Systemic Risk General Session Chair: Nils Detering, 1 - Credit Freezes, Equilibrium Multiplicity and Optimal Bailouts in Financial Networks Agathe Pernoud, Stanford University, Stanford, CA, United States We analyze how interdependencies between organizations in financial networks can lead to multiple possible equilibrium outcomes. A multiplicity arises if and only if there exists a certain type of dependency cycle in the network that allows for self-fulfilling chains of defaults. We provide necessary and sufficient conditions for banks’ solvency in any equilibrium. Building on these conditions, we characterize the minimum bailout payments needed to ensure systemic solvency, as well as how solvency can be ensured by guaranteeing a specific set of debt payments. We show that the minimum bailout problem is computationally hard, but provide an upper bound on optimal payments and show that the problem has intuitive solutions in specific network structures such as those with disjoint cycles or a core-periphery structure. 2 - Zooming in Distress Anomaly: Bankruptcy vs. Other Failures Xiaorui Zhu, University of Cincinnati, Cincinnati, OH, United States, Yuhang Xing, Yan Yu This paper reinvestigates the distress risk anomaly that financially distressed firms deliver abnormally low returns. We distinguish between bankruptcy and other- failure events and then utilizing the state-of-the-art adaptive Lasso variable selection method to identify predictors for these two types of risk. We obtain strikingly different predictors of bankruptcy and other-failure risk. In addition, both selected models gain better out-of-sample prediction performances than that of classical models in the literature. With the new risk measures, we find that the other-failure risk anomaly disappears while the anomalous return is persistently associated with the bankruptcy risk. 3 - When Do You Stop Supporting Your Bankrupt Subsidiary? Nils Detering, Santa Barbara, CA, 93106, United States Most global banks consist of dozens if not hundreds of subsidiaries. For corporations becoming a holding with subsidiaries has several advantages: It allows them to limit the spillover risk if one business line is in trouble or to defer taxable business income. For example a bank holding may naturally divide into subsidiaries based on location (i.e. Europe, US) and/or business lines (i.e. equity trading, fixed income trading). We consider a network of holding banks. We show that comprising a firms business activity in a holding can have both, positive or negative effects on systemic stability. We analyse to what extent voluntary support benefits society and/or the holding itself. We observe that an increased commitment of the holding to its subsidiaries can have both, positive and negative effects on systemic risk depending on the type of the holding. WD31 CC Room 208A In Person: Operations of Matching Markets General Session Chair: Vahideh Manshadi, Yale University, Quincy, MA, 02169-4688, United States 1 - Online Ranking Policies for Maximizing Engagement on Nonprofit Matching Platforms Akshaya Suresh, Yale University, New Haven, CT, United States, Vahideh Manshadi, Daniela Saban, Scott Rodilitz Nonprofit platforms that facilitate connections between volunteers and opportunities rely on their on-platform ranking engines as well as off-platform targeted promotion of opportunities to engage volunteers. In collaboration with VolunteerMatch, the largest of such platforms, we show that off-platform traffic constitutes a large portion of engagement but that opportunities enjoy disparate levels of such traffic. We develop ranking policies that effectively utilize off- platform traffic to maximize overall engagement and demonstrate their effectiveness by testing them on VolunteerMatch’s data. 2 - Two-Sided Assortment Optimization Ignacio Rios, Assistant Professor, The University of Texas at Dallas, Richardson, TX, United States We consider a two-sided market mediated by platform, where agents on each side of the market see a subset of profiles in each period. Matches are generated if two users mutually like each other, possibly on different periods. The goal of the platform is to maximize the expected number of matches generated. We model this problem as a dynamic program, we analyze its properties, and we provide performance guarantees for some particular cases of special interest.

171