2016 INFORMS Annual Meeting Program

TC46

INFORMS Nashville – 2016

TC46 209B-MCC Empirical and Data-Driven Studies in Revenue Management and Pricing Sponsored: Revenue Management & Pricing Sponsored Session Chair: Jun Li, Ross School of Business, University of Michigan, Ann, MI, United States, junwli@umich.edu Co-Chair: Serguei Netessine, INSEAD, Singapore, Singapore, serguei.netessine@insead.edu 1 - Compete With Many: Price Competition In High Dimensional Space We study price competition in markets with a large number (in the magnitude of hundreds or thousands) of potential competitors. We propose a new instrument variable approach to address simultaneity bias in high dimensional variable selection problems. The novelty of the idea is to exploit online search and clickstream data to uncover demand shocks at a granular level, with sufficient variations both over time and across competitors. We apply this data-driven approach to study the New York City hotel market. 2 - Randomized Markdowns And Online Monitoring Ken Moon, The Wharton School, Philadelphia, PA, United States, kenmoon@wharton.upenn.edu, Kostas Bimpikis, Haim Mendelson Using data tracking customers of a North American retailer, we present empirical evidence that monitoring products online associates with successfully obtaining discounts. We develop a structural model of consumers’ dynamic monitoring to find substantial heterogeneity, with consumers’ opportunity costs for an online visit ranging from $2 to $25 in inverse relation to their price elasticities. We show implications for retail operations and also discuss targeting customers with price promotions using their online histories. 3 - A Nonparametric Approach To Learning Mixture Models Srikanth Jagabathula, NYU Stern, sjagabat@stern.nyu.edu, Lakshminarayanan Subramanian, Ashwin Venkataraman We consider the problem of learning a mixture model, where the number of mixture components is learned directly from the data. Our framework applies to any mixture model, but we specialize our techniques to learn the mixture of multinomial (MNL) models. We pose the learning problem as that of minimizing a loss function (likelihood, squared-loss, etc.) over the data and model parameters. Common formulations result in a non-convex problems. We overcome this through a novel reformulation that converts the problem into a semi-infinite convex program. We then apply conditional-gradient techniques to solve the convex program. We validate our methods theoretically and empirically. 4 - Leveraging Inventory In Profit Maximization For Personalized Online Bundle Pricing Recommendation Anna M Papush, Massachusetts Institute of Technology, 1 Amherst Street, Cambridge, MA, 02139, United States, apapush@mit.edu, Pavithra Harsha, Georgia Perakis E-commerce has been vastly growing in popularity over the past decade. It has been capturing increasing proportions of the retail market. Thus gaining the competitive edge in this sector is of utmost importance to any firm’s success. This work presents a model for providing relevant product recommendations at personalized prices while leveraging knowledge of inventory at-risk for markdowns and maximizing retailer profits. We demonstrate practical applications through implementation on actual e-tailer data, as well as establishing performance guarantees relative to an optimal offline approach. TC47 209C-MCC Choice-based Demand and Strategic Consumers Sponsored: Revenue Management & Pricing Sponsored Session Chair: Gustavo Vulcano, New York University, 44 West Fourth St, Suite 8-76, New York, NY, 10012, United States, gvulcano@stern.nyu.edu 1 - Predicting Individual Customer Responses To Product Promotions Dmitry Mitrofanov, NYU, dm3537@stern.nyu.edu, Srikanth Jagabathula, Gustavo Jose Vulcano We consider the problem of predicting an individual customer’s response to a product promotion using historical purchase transactions data, tagged by the customer id. The problem is challenging because of the limited number of observations available for each individual. To extract the signal from the limited data most efficiently, we model each individual through a partial-order consisting Jun Li, Ross School of Business, University of Michigan, junwli@umich.edu, Serguei Netessine, Sergei Koulayev

of weak and strong preferences (A is strongly preferred to B if non-promoted A is purchased over promoted B). We calibrate an MNL model over the partial orders and quantify its prediction power on out-of-sample transactions. Then, we use this information to optimize personalized promotions. 2 - The Heteroscedastic Exponomial Choice Model Aydin Alptekinoglu, Penn State, aydin@psu.edu, John H Semple We develop analytical properties of the Heteroscedastic Exponomial Choice (HEC) model and demonstrate its estimation using a household panel data of grocery purchases. The HEC model compares quite favorably to MNL in out-of-sample prediction. 3 - Assortment Optimization With Product Costs And Constraints Sumit Kunnumkal, ISB, sumit_kunnumkal@isb.edu, Victor Martinez de Albeniz We consider the assortment optimization problem under the MNL model with product fixed costs and constraints. We propose a new method to obtain an upper bound on the optimal expected profit. We show that our method is tractable and has provable performance guarantees for some common types of assortment constraints. 4 - Optimal Pricing In Continuous Time José Correa, University of Chile, Santiago, Chile, correa@uchile.cl We consider continuous time pricing problems with strategicconsumers that arrive over time. By combining ideas from auctiontheory and recent work on pricing with strategic consumers we derivethe optimal continuous time pricing scheme in some situations. Ournovel approach is based on optimal control theory and is well suitedfor numerical computations. 210-MCC Social Media Analysis IV Invited: Social Media Analytics Invited Session Chair: Anandasivam Gopal, University of Maryland, Smith School of Business, College Park, MD, 11111, United States, agopal@rhsmith.umd.edu 1 - Extraction Of Adverse Events From Social Media Lina Zhou, University of Maryland-Baltimore County, 1000 Hilltop Circle, Baltimore, MD, 21250, United States, zhoul@umbc.edu, Yin Kang Adverse events have significant impacts on patients’ safety. Understanding patient reports of adverse events in social media remains a significant research challenge. We proposed new methods based on syntactic dependency relations to extract adverse events. The experiment results demonstrate that our proposed methods improve the extraction performance across all data sets in terms of both precision and recall. 2 - When It Rains, It Pours: Effect Of Social Media On Stock Price Behavior During Firm Crises Soo Jeong Hong, Michigan State University, 404 Wilson Rd, Room 311, East Lansing, MI, 48824, United States, hongsoo3@msu.edu, Kwangjin Lee We examine the effect of social media usage on the capital market under a firm crisis situation. Focusing on consumer product recalls, we investigate the factors which determine individuals’ information sharing decisions in social media. We also analyze when recall information shared via social media can exacerbate negative market reactions. 3 - Networked Pattern Recognition Frameworks For Understanding And Detecting Future Terrorism Threats Salih Tutun, Turkish Military Academy, and Binghamton University, 4400 Vestal Pkwy E, Binghamton, NY, 13850, United States, slh.tutun@gmail.com, Mohammad Khasawneh The challenge of governments are how to track threats, since terrorists have learned how to avoid unsecured communications, such as social media. This research is proposed as a new framework that will better understand the characteristics of future suicide attacks by analyzing the relationship among the attacks. It is also proposed as a new unified detection framework that applies pattern classification techniques to network topology to detect terrorist activity. The finding results can potentially use to propose reactive strategies thus enabling precautions to be taken against future attacks. TC48

318