Informs Annual Meeting 2017

SA38

INFORMS Houston – 2017

SA36

Columbia Pike, Apt 811, Silver Spring, MD, 20904, United States, wzhang@rhsmtih.umd.edu, Kostas Bimpikis, Wedad J.Elmaghraby, Ken Moon Platforms can obtain sizable returns by operationally managing their market thickness, i.e., the availability of supply-side inventory. Using data from a natural experiment on a major B2B auction platform specializing in the $424 billion secondary market for liquidating retail merchandise, we find that thickening the platform’s market by consolidating the ending times of auctions to certain weekdays substantially increases its revenue by 6.5%, due primarily to the bidders’ participation frictions. We study two complementary design levers to calibrate and control the platform’s market thickness in the face of complex demand-side decision making: listing policy and a recommendation system. 2 - Examining Impacts of Clinical Practice Variation on Operational Performance: Implications for Bundled Payment Reform Models Seokjun Youn, Texas A&M.University, 320 Wehner Building, 4217 Texas A&M.University, College Station, TX, 77840-4217, United States, syoun@mays.tamu.edu, Gregory R. Heim, Subdoha Kumar, Chelliah Sriskandarajah Motivated by bundled payment policies that aim to reduce practice variation, this study examines whether and how lower variations in clinical practice relate to hospital operational performance. Analyzing six years of data from hospitals in NY and FL states, we provide our empirical findings. We believe better measurement and understanding of practice variation’s antecedents and consequences can lead managers and policy-makers to better design bundled payment reform models, the target opportunity of which is to reduce waste by decreasing variabilities in care-delivery processes. 3 - Linking Delay Announcements, Abandonment, and Staffing: A Behavioral Perspective Eric Michael Webb, eric.michael.webb@gmail.com, Kurt Bretthauer, Qiuping Yu Using field data from a call center, we study the behavioral determinants of customers’ queue abandonment decisions in the presence of delay announcements. We relax traditional behavioral assumptions on how customers should behave. Customers exhibit loss aversion with respect to time losses in queue, becoming much more likely to abandon if forced to wait longer than expected. Customers also exhibit a decreasing base hazard rate. Longer announced wait times increase customer abandonment behavior and customers are often strategic in their arrival decisions. By accounting for our insights, firms may significantly reduce their staffing levels without increasing customers’ overall waiting time. 352C Algorithms and Heuristics for Multiobjective Mixed-integer Optimization Sponsored: Multiple Criteria Decision Making Sponsored Session Chair: Daniel Jornada, Ford Motor Company, 15403 Commerce Drive South, Dearborn, MI, 48120, United States, danieljornada@gmail.com 1 - Bi-objective Mixed Integer Linear Programming for Managing Building Clusters with a Shared Electrical Energy Storage Rui Dai, University of South Florida, Tampa, FL, United States, ruidai@mail.usf.edu Building clusters allow buildings to collaborate and reduce their operational cost. However, since each building is treated as an independent decision maker, fairness havs to be addressed in these collaborative environments. We address these issues on building clusters with an electrical energy storage shared between two buildings with deterministic demand. We introduce three strategies, and develop a bi-objective mathematical formulation for each strategy with several linearization techniques such as piecewise McCormick relaxation. Experiment results demonstrate the efficacy of our proposed linearization techniques, and compare all three strategies in terms of fairness and freedom. 2 - The Box-Line Algorithm for Mixed Integer Biobjective Optimization Diego Pecin, Georgia Tech, Atlanta, GA, United States, diegopecin@gmail.com Recent years have seen a surge of new algorithms for solving multiobjective integer programs, especially for the pure integer case. The presence of continuous variables presents significant challenges to discovery of the nondominated frontier, and development of algorithms for the mixed integer case have lagged. Here, we present a new algorithm for mixed integer problems with two objectives, which generalizes the Balanced Box Method (BBM) for pure integer programs. The computational performance of the Box-Line Algorithm is compared with that of existing methods on benchmark problems, in terms of its overall run-time, and its ability to approximate the frontier if terminated early. SA38

351F Service Science Analytics Sponsored: Service Science Sponsored Session Chair: Suchithra Rajendran, PhD, Pennsylvania State University, State College, PA, 16802, United States, sur205@psu.edu 1 - Minimizing Completion Time Variance in Service Industries using Heuristic Approaches Chandrasekharan Rajendran, IIT.Madras, Dept. of Mgmt. Studies, IIT Madras-Chennai, Chennai, 600040, India, crajiitm@gmail.com The problem of scheduling customers or jobs in a service system or in a manufacturing system is an important operational issue. Minimizing the completion-time variance (CTV) of customers is an important performance measure when customers require an uniform treatment, i.e., each customer spends approximately the same time in the system or waits for service as every other entity. In this work, an existing branch-and-bound algorithm is first analyzed for minimizing the CTV and a correction on the lower bound is presented. Based on this, a simple heuristic to solve the problem is proposed. 2 - Determining Risk Type of Customers at Financial Institutions using Machine Learning Algorithms Suchithra Rajendran, Pennsylvania State University, 265 Blue Course Drive, Apt 3E, State College, PA, 16803, United States, sur205@psu.edu, Sharan Srinivas Financial institutions are constantly facing the problem of approving loans to high-risk customers. It has been reported in the literature that there are several financial institutions becoming bankrupt due to loan defaulting. In this research, five machine learning approaches are developed to classify the customers as high risk and low risk based on several parameters such as age, salary and past history on defaulting. The methods are compared using several performance measures such as sensitivity, specificity and accuracy to determine the best performing rule. 3 - An Integrated Supplier Selection and Lot Sizing Policy Considering Total Quantity Discount and Quality Constraint Sang Jin Kweon, PhD, Pennsylvania State University, Industrial and Manufacturing Engineering, 232 Leonhard Building, University Park, PA, 16802, United States, rossa1984@gmail.com, Xin Li, Jose Antonio Ventura, Barbara Venegas, Seong Wook Hwang This paper studies a joint supplier selection and lot sizing problem with quality constraint. Suppliers are capacitated and offer total quantity discount (TQD). Under TQD, the product unit price depends on the total order quantity (TOQ) for the supplier over a certain time period. Order quantities can have integer or continuous values. For each case, given a supplier and the TOQ for it, a procedure to derive the optimal order quantities and number of orders is developed. A near- optimal algorithm that combines Lagrangian relaxation and dynamic programming is proposed to solve the problem. 4 - Preference Based Scheduling in a Healthcare Provider Network Deepak Agrawal, Pennsylvania State University, 310 Leonhard Healthcare is delivered by a complex network of physicians, clinic facilities and patients with different priorities and preferences. Traditional appointment scheduling systems do not consider the network effect. Motivated by this, we study the advanced scheduling problem with multiple clinic locations and physicians where patients can visit any clinic location and provider of their preference. We develop theoretical bounds on the performance gap with respect to the best achievable policy and propose a performance improvement heuristic to solve the problem. Further, we conduct numerical experiments to show that our proposed model outperforms the scheduling policies used in practice. 352B Best IBM Service Science Student Paper Award Competition I Sponsored: Service Science Sponsored Session Chair: Robin Qiu, Pennsylvania State University, Malvern, PA, 19355, United States, robinqiu@psu.edu Co-Chair: Aly Megahed, IBM Research-Almaden, San Jose, CA, 95123, United States, aly@gatech.edu 1 - Managing Market Thickness in Online B2B Markets Wenchang Zhang, Robert H.Smith School of Business, 12001 Old Building, University Park, PA, 16802, United States, dua143@psu.edu, Guodong Pang, Soundar Kumara SA37

15