2016 INFORMS Annual Meeting Program

WD47

INFORMS Nashville – 2016

WD45 209A-MCC Network Economics II Sponsored: Simulation Sponsored Session Chair: Bruno Tuffin, INRIA, TBD, Rennes Cedex, France, bruno.tuffin@inria.fr Co-Chair: Patrick Maillé, Telecom Bretagne, 2, rue de la Cha^taigneraie, Cesson Sévigné, 35576, France, patrick.maille@telecom-bretagne.eu 1 - On Revenue-oriented Content Delivery Networks And Their Impact On Net Neutrality Patrick Maillé, Telecom Bretagne, Rennes, 35510, France, patrick.maille@telecom-bretagne.eu, Gwendal Simon, Bruno Tuffin We investigate the decisions made by a CDN actor willing to maximize revenue through the management of its cache servers. Through simple models, we highlight that revenue-oriented management policies can affect the user- perceived quality of experience, impacting the competition among content providers (and also among network access providers) in favor of the incumbent. Since this goes in the opposite direction to the one aimed by net neutrality proponents and it seems that CDNs are not discussed much in the net neutrality debate, we wonder about the need for a definition of what a “neutral” CDN should look like. 2 - An Optimization Approach For Long-term Network Planning With Protection Constraints Nicolas Stier-Moses, Facebook, Menlo Park, CA, United States, nicostier@yahoo.com, Josue Kuri We present an optimization approach for strategic, long-term network planning. We forecast supply (network assets), demand (network traffic) and possible failures, and employ a robust optimization approach to optimize which assets to use and how much capacity must be turned up in each of them. A key element of this model is its dynamic nature which allows us to consider inter-temporal constraints (e.g., turned up capacity is monotone over time) and the amortization of fixed costs. 3 - Profit, Welfare, And Consumer Surplus Implications Of Sponsored Data Plans Jialin Song, University of Illinois at Urbana-Champaign, 117 Transportation Building, 104 S Mathews Ave, Urbana, IL, 61801, United States, jsong83@illinois.edu, Qiong Wang Major Mobile Service Providers are now offering sponsored data plans that allow Content Providers to pay for the access of their contents by end users. How such practice affects profit, social welfare, and consumer surplus is a critical question in the network neutrality debate. We address this issue from the perspective of commodity bundling: without a sponsored data plan, users purchase data blocks to access all contents, which corresponds to pure bundling; sponsored data plans separate some contents from the bundle. We develop a two-stage model, involving both Nash bargaining solution and Nash equilibrium, to analyze and compare the two situations. Revenue Management and Marketing Sponsored: Revenue Management & Pricing Sponsored Session Chair: Denis Saure, University of Chile, Beaucheff 851, Santiago, Chile, dsaure@dii.uchile.cl Co-Chair: Juan Pablo Vielma, Massachusetts Institute of Technology, 30 Memorial Dr, Cambridge, MA, 02142, United States, jvielma@mit.edu 1 - Ellipsoidal Methods For Choice-based Conjoint Analysis Denis Saure, Universidad de Chile, Republica 701, Santiago, 8370439, Chile, dsaure@dii.uchile.cl, Juan Pablo Vielma In this talk we introduce a variant of the polyhedral method by Toubia, Hauser and Simester (2004) that uses ellipsoids instead of polyhedral. This change allows the method to (1) include approximate Gaussian priors on the parameters, (2) explicitly consider respondent error, and (3) perform quick approximate posterior updates whose quality nearly matches a full Bayesian update. We also introduce a practical question selection method that is optimal with respect to the D- efficiency criterion for one question, and leads to an extremely effective one-step look-ahead policy for multiple questions. WD46 209B-MCC

2 - Capturing Multitaste Preferences: A Machine Learning Approach Daria Dzyabura, New York University, ddzyabur@stern.nyu.edu, Liu Liu In diverse product categories, a consumer’s preferences may include several tastes. For example, one may enjoy cooking American and Chinese recipes, with different specific criteria for each. We propose a model that allows for multiple tastes and an efficient estimation algorithm. In a numerical study, we simulate multi-taste consumers and demonstrate the proposed algorithm accurately recovers parameters, while benchmark models underfit. We test the algorithm on recipe texts, after extracting attributes from recipe text using text mining. We achieve significant improvements in prediction over single-taste benchmarks. 3 - Estimating Customer Spillover Learning Of Service Quality: A Bayesianhierarchical Andres I Musalem, Universidad de Chile, Beauchef 851, Santiago, 8370456, Chile, amusalem@duke.edu, Yan Shang, Jeannette Song We propose a Bayesian framework for estimating customer “spillover learning,” — the process by which customers’ learn from previous experiences of similar but not necessarily identical services. We apply our model to a data set containing shipping and sales historical records provided by a world-leading third-party logistics company. WD47 209C-MCC Various Pricing Topics for Revenue Management Sponsored: Revenue Management & Pricing Sponsored Session Chair: Elcin Ergin, McGill, 3465 Hutchison Street, Apt 905, Montreal, QC, H2X 2G3, Canada, elcin.ergin@mail.mcgill.ca 1 - Better Late Than Now: Delayed Vs. Instantaneous Price Discounts With Repeat Customers Monire Jalili, PhD Student, University of Oregon, 1455 East 25th Ave, Eugene, OR, 97403, United States, mjalili@uoregon.edu, Michael Pangburn In this paper, we contrast the delayed versus instantaneous discounting policies in a repeat purchase setting with rational and forward-looking consumers. We first establish that if consumer spending is consistent over time, then there is no benefit to the firm (or consumers) from delayed discounts. With varied spending, we prove that when the firm can target individual consumers with their optimal discount percentage, delaying discounts increases profits only for a limited range of transition shoppers. However, when the same discount percentage applies to all customers, delayed discounting outperforms the instantaneous discounting, thus motivating the prevalence of this policy in practice. 2 - Competitive Pricing With Stockouts And Satisficing Customers Varun Gupta, Penn State Erie, The Behrend College, 5101 Jordan Rd, Burke 281, Penn State Erie, The Behrend College, Erie, PA, 16563, United States, vxg15@psu.edu, Metin Cakanyildirim Stockouts for high inventory turnover products lead to loss of sales as customers may substitute their preferred product (stocked out) with another product (available). We study single period equilibrium prices for competing retailers selling to satisficing customers with stockout-based substitution under lost sales and backorders. 3 - Pricing Decisions In Fast Fashion Retailing Using Discrete Choice Dynamic Programming Model Elcin Ergin, McGill, 3465 Hutchison Street, Apt 905, Montreal, QC, H2X 2G3, Canada, elcin.ergin@mail.mcgill.ca, Mehmet Gumus We study the pricing decisions in fast-fashion retailing firms under a forward- looking setting utilizing the dynamic nature of the problem. In this context, we consider a discrete choice dynamic programming model to estimate the optimal pricing decisions throughout the life cycle of a product. We develop decomposition approaches based on different functional forms and assess their performances in terms of computational complexity and objectives of the problem on a large real-life dataset taken from a fast-fashion company.

469