Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD22

1 - The Rise of Ship-to-store: Theoretical and Empirical Analyses of Its Impact on Online Sales Necati Ertekin, Santa Clara University, Department of OM&IS, Leavey School of Business, Santa Clara, CA, 95053, United States, Mehmet Gumus, Mohammad Nikoofal In this paper, we first develop a customer choice model to generate theoretical predictions regarding the impact of ship-to-store (STS) on online sales for two types of products, namely online-exclusive products (i.e., products available only online) and hybrid products (i.e., products available both online and offline), as well as on overall online sales. Next, we empirically test our predictions using data from an omnichannel retailer that launched STS at two different brand names. We find that, due to STS, online sales decrease at one brand name and increase at the other. We explain this controversy with our theoretical model and the extensive empirical analysis. 2 - An Empirical Analysis of Intra Firm Product Substitutability in Fashion Retailing Elcin Ergin, McGill University, 1001 Sherbrooke St W, Room 521, Montreal, QC, H3A 1G5, Canada, Mehmet Gumus, Nathan Yang Our study investigates the impact of product shortages on sales in neighboring outlets using novel data from a fast fashion retailing chain. Our analysis reveals that sales for an item at a focal store increases when that same item experiences a stock out in neighboring store. We show that these product substitutability patterns across stores vary depending on the location type. We also demonstrate in a forecasting exercise that information about stock outs in neighboring outlets has value for prediction for certain periods of the product’s life cycle. Finally, we investigate the revenue impact of making inventory allocation decisions using the neighboring outlets’ stock out information. 3 - Generating Realistic Customer Purchase Baskets Using Generative Adversarial Networks Saibal Ray, McGill University, Desautels Faculty of Management, 1001 Sherbrooke Street W, Montreal, QC, H3A 1G5, Canada, Thang Doan, Brian Keng This project uses the purchase history from loyalty member card at the basket level (i.e., all items bought by a customer during a particular trip) for customers under the loyalty program of a chain drug store to develop a model that can generate realistic future “customer shopping list (basket) using the novel machine learning technique of Generative Adversarial Networks (GAN). Our first objective in this project is to find a representation/embedding of customers and the baskets they buy by analyzing the transaction histories. Once the above task has been accomplished, we will build a customer simulator that will generate realistic future customers and their purchase baskets for the next 4 weeks 4 - Visual Listening In: Extracting Brand Image Portrayed on Social Media Liu Liu, New York University, New York, NY, United States, Daria Dzyabura, Natalie Mizik Images are on their way to surpassing text as the medium of choice for social conversations. In this paper, we propose a ``visual listening in” approach to measuring how brands are portrayed on social media (Instagram), by mining visual content posted by users. We first train image classifiers of brand attributes via SVM and Convolutional neural networks. Then we apply the classifiers to brand-related images on social media. We study 56 brands in the apparel and beverages categories, and compare their portrayal in consumer-created images with images on the firm’s official Instagram account, as well as with consumer brand perceptions measured in a national brand survey to derive managerial insights. n MD22 North Bldg 130 Consumer Oriented Operations Models Sponsored: Revenue Management & Pricing Sponsored Session Chair: Mingcheng Wei, University at Buffalo, Buffalo, NY, 14260, United States 1 - Optimal Pricing in Online Marketplaces Xuanming Su, University of Pennsylvania, The Wharton School, 3730 Walnut Street, Philadelphia, PA, 19104, United States We develop a pricing model for online marketplaces that match demand from customers with supply from firms. We consider prices between customers, firms, workers, and the marketplace, and then relate our results to current practice.

n MD20 North Bldg 129A Learning and Platforms Sponsored: Revenue Management & Pricing Sponsored Session Chair: Ramesh Johari, Stanford University, Stanford, CA, 94305-4121, United States 1 - A Conditional Gradient Approach for Nonparametric Estimation of Mixing Distributions Ashwin Venkataraman, Harvard University, Cambridge, MA, United States, Srikanth Jagabathula, Lakshminarayanan Subramanian A key challenge in estimating mixture models is that the mixing distribution is often unknown and imposing apriori parametric assumptions can lead to model misspecification issues. We propose a new methodology for nonparametric estimation of the mixing distribution. We formulate the likelihood-based estimation problem as a constrained convex program and our key contribution is applying the conditional gradient (aka Frank-Wolfe) algorithm to solve this convex program, showing that it iteratively generates the support of the mixing distribution. We show that our estimator is robust to different ground-truth mixing distributions and outperforms the EM benchmark in two case studies on real data. 2 - Bandit Learning with Positive Externalities Virag Shah, Stanford University, Mail Code: 4026, Management Science and Engineering, Stanford, CA, 94305, United States, Jose Blanchet, Ramesh Johari In many platforms the future user arrivals are likely to have preferences similar to users who were satisfied in the past. In other words, arrivals exhibit positive externalities. We study multiarmed bandit (MAB) problems with positive externalities. We show that the self-reinforcing preferences may lead standard benchmarks such as UCB to exhibit linear regret. We develop a new algorithm, Balanced Exploration, which explores arms carefully to avoid suboptimal convergence of arrivals before sufficient evidence is gathered, and an adaptive variant which successively eliminates suboptimal arms. We analyze their regret and establish optimality by showing that no algorithm can perform better. 3 - Incentivizing Exploration by Heterogeneous Users Peter Frazier, Cornell University, School of Operations Research, and Information Engineering, Ithaca, NY, 14853, United States, Bangrui Chen We consider incentivizing exploration with heterogeneous agents. In this problem, bandit arms provide random attribute vectors from unknown fixed distributions over which agents have heterogeneous utilities. Selfish myopic agents arrive sequentially, observe past pulls, and pull the arm with the largest expected utility. Agents may be incentivized to pull underexplored arms through payments. We design an incentivization algorithm whose expected cumulative regret and payment are constant in the time horizon T, when agent types are finite and all all arms are preferred by some agents. This contrasts with log(T) regret in the standard MAB. Succinctly, heterogeneity provides free exploration. 4 - Stochastic Bandits Robust to Adversarial Corruptions Thodoris Lykouris, PhD Candidate, Cornell University, 107 Hoy Road, Gates 336, Ithaca, NY, 14853, United States, Vahab Mirrokni, Renato Paes Leme We introduce a new model of stochastic bandits with adversarial corruptions which captures settings where most of the input follows a stochastic pattern but a fraction of it is adversarially changed to trick the algorithm, e.g. click fraud, fake reviews, and email spam. The goal of this model is to encourage the design of bandit algorithms that (i) work well in mixed adversarial and stochastic models, and (ii) whose performance deteriorates gracefully as we move from fully stochastic to fully adversarial models. We provide an algorithm whose performance gracefully degrades with the total corruption the adversary injected in the data (while being agnostic to it) and a corresponding lower bound. n MD21 North Bldg 129B Data-driven and Machine Learning Research in Retail Management Sponsored: Revenue Management & Pricing Sponsored Session Chair: Mehmet Gumus, McGill University, Montreal, QC, H3A 1G5, Canada Co-Chair: Saibal Ray, McGill University, Montreal, QC, H3A 1G5, Canada

221