2015 Informs Annual Meeting

TA43

INFORMS Philadelphia – 2015

3 - Screening for Hepatocellular Carcinoma: A Restless Bandit Model Elliot Lee, University of Michigan, 1205 Beal Ave, Ann Arbor, MI, 48109, United States of America, elliotdl@umich.edu, Mariel Lavieri, Michael Volk Currently, all patients at risk for hepatocellular carcinoma (HCC) are screened every six months. Recent medical discoveries have found a correlation between a biomarker measured at each screening, and his/her risk of developing HCC. We model the problem of simultaneously learning while allocating a limited number of screening resources across a population as a restless bandit model. We prove several structural properties of this problem, and ultimately derive a corresponding optimal policy. 4 - Enhancing FDA’s Decision Making using Data Analytics Vishal Ahuja, Southern Methodist University, P.O. Box 750333, TX, United States of America, vahuja@smu.edu, John Birge Existing FDA surveillance methods are based on voluntary reporting or meta- analysis primarily geared towards identifying new/unknown adverse events. We propose a statistically robust and evidence-based empirical approach that focuses on evaluating specific drug-related adverse outcomes to aid in the FDA decision- making. We demonstrate our approach using a controversial black box warning. Based on a large dataset from the Department of Veterans Affairs, we find that the warning was not warranted. TA43 43-Room 103A, CC Measurement and Optimization in Online Advertising Sponsor: Revenue Management and Pricing Sponsored Session Chair: Omar Besbes, Professor, Columbia University, Graduate School of Business, New York, NY, 10027, United States of America, ob2105@columbia.edu Co-Chair: Vineet Goyal, Associate Professor, Industrial Engineering and Operations Research, Columbia University, 500 West 120th Street, 304 Mudd, New York, NY, 10027, United States of America, vgoyal@ieor.columbia.edu Co-Chair: Garud Iyengar, Columbia University, S. W. Mudd 314, 500W 120th Street, New York, NY, United States of America, garud@ieor.columbia.edu 1 - Advertiser Revenue Versus Consumer Privacy in Online Advertising Vibhanshu Abhishek, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, vibs@andrew.cmu.edu, Arslan Aziz, Rahul Telang Increasing concerns around consumer privacy have questioned the value of targeted advertising. In this paper we quantify the value of privacy-intrusive information in targeted advertising. Using individual level browsing/purchase data we find that using more privacy-intrusive information increases the accuracy of prediction of purchases, but at a decreasing rate. Targeted advertising is also effective in increasing purchase probability. In our data, restricting cookies reduces sales by 14%. 2 - Learning and Optimizing Reserve Prices in Repeated Auctions Hamid Nazerzadeh, University of Southern California, Bridge Memorial Hall, 3670 Trousdale Parkway, Los Angeles, 90089, A large fraction of online advertisements are sold via repeated second price auctions. The reserve price is the main tool for the auctioneer to boost revenues. However, the question of how to effectively set these reserves remains essentially open from both theatrical and practical perspectives. The main challenge here is that using previous bids to learn reserves could lead to shading of bids and loss of revenue. I’ll present incentive compatible near-optimal leaning algorithms in this context. 3 - Advertising on a Map Sergei Vassilvitskii, Google, 111 8th Avenue, New York, United States of America, sergei@cs.stanford.edu We study the mechanism design problem for advertising on a map. Unlike traditional search advertising where there is a linear order on the slots, no such structure exists in the case of a map. We begin with a model of the setting, noting that the utility of an ad is discounted by the presence of competing businesses nearby and its position in the set of ads ordered by distance from the user. We present simple, approximately welfare maximizing allocation schemes with good incentive properties. United States of America, hamidnz@marshall.usc.edu, Yash Kanoria, Renato Paes Leme, Afshin Rostamizadeh, Umar Syed

4 - Attribution in Online Advertising under Markov Browsing Models Antoine Desir, Columbia University IEOR department, 500 West 120th Street, Mudd 315, New York, NY, 10027, United States of America, ad2918@columbia.edu, Vineet Goyal, Garud Iyengar, Omar Besbes Web viewers are exposed to multiple ads across different websites before they potentially make a purchase (leading to a conversion). In turn, a key question facing the online advertising industry is that of attribution. We analyze attribution based on Shapley values. While intractable in general, we provide computationally tractable approximations to Shapley values under a general Markov chain customer browsing behavior model and compare this attribution to heuristics commonly used in practice. TA44 44-Room 103B, CC Pricing Issues in Revenue Management Sponsor: Revenue Management and Pricing Sponsored Session Chair: Rene Caldentey, NYU, 44 W 4th St, New York, NY, 10012, United States of America, rcaldent@stern.nyu.edu Co-Chair: Ying Liu, Stern School of Business, New York University, 44 West 4th Street, KMC 8-154, New York, NY, 10012, United States of America, yliu2@stern.nyu.edu 1 - Incorporating Online Customer Ratings in Pricing Decisions Marie-claude Cote, Manager, Data Science, JDA Software - Innovation Labs, 4200 Saint-Laurent #407, Montréal, QC, H2W 2R2, Canada, Marie-Claude.Cote@jda.com, Philippe Tilly, Nicolas Chapados Research have demonstrated that online customer ratings have a huge impact on the decision to choose a product. In hospitality, where the product is a hotel room for a length of stay, customers consult an increasing number of reviews prior to booking.We will describe an approach to automatically incorporate online user rating impact in the pricing decisions of a hospitality revenue management system. 2 - On the Equivalence of Quantity Pre-commitment and Cournot Games Amr Farahat, Washington University in St. Louis, One Brookings Drive, St. Louis, MO, 63104, United States of America, farahat@wustl.edu, Hongmin Li, Tim Huh We establish sufficient conditions under which Cournot outcomes solve quantity- followed-by-pricing games. Kreps and Scheinkman (1983) established this connection for homogeneous product duopolies and Friedman (1988) for certain differentiated product oligopolies under restrictive assumptions. Our research provides conditions for more general differentiated product settings, including multinomial logit models. 3 - Pricing Policies for Perishable Products with Demand Substitution Ying Liu, Stern School of Business, New York University, 44 West 4th Street, KMC 8-154, New York, NY, 10012, United States of America, yliu2@stern.nyu.edu, Rene Caldentey We study a monopolist’s optimal dynamic pricing policy for a family of substitute perishable products. Customers arrive to the market according to an exogenous stochastic process, each with a budget constraint. Upon arrival, each customer first makes a decision among subfamilies that are differentiated by quality, and then selects among horizontally differentiated products within the subfamily. We characterize the optimal pricing policy and study the asymptotic approximation. 4 - Optimal Time and Price of Dynamic Upgrade Xiao Zhang, PhD Candidate, University of Texas at Dallas, Richardson, TX, United States of America, xxz085020@utdallas.edu, Ozalp Ozer Upgrade, a strategy used in travel industry to balance the supply-and-demand mismatches among products of different quality levels, is usually offered either at the booking time or the consumption time. We study a revenue management problem of a firm which sells two products at fixed prices and offers upgrade options anytime when necessary. The optimal policy specifies the time and price of the upgrade option, and how many existing customers should be offered this option.

272