Informs Annual Meeting 2017

SB27

INFORMS Houston – 2017

SB27

2 - Optimal Quantity Caps in Discriminatory Price Auctions with Resale Justin Burkett, Wake Forest University, 1834 Wake Forest Road, Department of Economics, Winston-Salem, NC, 27106, United States, burketje@wfu.edu, Brian Baisa We present a model of a discriminatory price auction in which a large bidder competes against many small bidders, followed by a post-auction resale stage in which the large bidder is endogenously determined to be a buyer or a seller. We extend first-price auction with resale results to this setting and give a tractable characterization of equilibrium behavior. We use this to study the policy of capping the amount that may be won by large bidders in the auction. Despite being used in the US Treasury auction among others, this policy has received little attention in the auction literature. 3 - Optimal Auction Design in the Presence of Competition Mallesh Pai, Rice University, Department of Economics, MS-22, Rice University P.O. Box 1892, Houston, TX, 77251-1892, United States, mallesh.pai@rice.edu I study the design of the optimal auctions in the presence of competition. Using the theory of mathematical programs with equilibrium constraints, I study the problem of a seller whose competitors have committed to using second price auctions. I provide necessary conditions on the distribution of buyer values for this seller’s optimal auction to be a second price auction with a reserve. The conditions are strict weakenings of Myerson’s increasing virtual value condition. 4 - Sample Complexity of Multi-item Profit Maximization Tuomas W. Sandholm, Carnegie Mellon University, Gates Center for Computer Science, Pittsburgh, PA, 15213, United States, sandholm@cs.cmu.edu, Maria-Florina Balcan, Ellen Vitercik In multi-item settings, it is typically unrealistic to assume the designer has a prior over buyers’ valuations. We analyze revenue-maximizing sales mechanisms when the designer has samples from the prior, introduced by Likhodedov and Sandholm in AAAI-04. We provide generalization guarantees that bound the difference between profit on the sample and profit over the prior. Our overarching theorem uses Rademacher complexity to measure intrinsic complexity of widely-studied single- and multi-item auction classes. It also applies to linear and non-linear single- and multi-item pricing. Furthermore, it enables one to determine the optimal complexity in mechanism hierarchies. 350E New Methods and Algorithms in Artificial Intelligence Sponsored: Artificial Intelligence Sponsored Session Chair: Bradley Walls, University of Arizona, Tucson, AZ, 8, United States, blwalls@email.arizona.edu 1 - Machine Learning Methods to Tease Out Hidden Patterns in Hierarchical Data We use Machine Learning methods to tease out hidden patterns in hierarchical data and forecast an outcome characterized by high dimensionality. Our methods outperform comparable statistical models and are robust in dealing with sparse data. We offer insights on extending the technique problem domains were the outcome is based on a model of effects characterized by high dimensionality. 2 - Optimal Regression Trees Jack Dunn, Massachusetts Institute of Technology, 24 Hardwick St, Unit 1, Cambridge, MA, 02141, United States, jackdunn@mit.edu, Dimitris Bertsimas Decision tree methods are widely used for regression problems, but one must choose between methods that are interpretable (such as CART) and those with high accuracy (random forests or boosting). We introduce a new approach for constructing Optimal Regression Trees based on mixed-integer optimization, and develop high-performance heuristics for these problems that offer significant improvements over traditional greedy approaches and run in comparable times. We show in a collection of synthetic and real-world instances that our Optimal Regression Trees improve substantially over CART and related methods such as Random Forests, delivering high-accuracy solutions that are interpretable. SB29 Praveen Pathak, University of Florida, 8325 SW. 16th Place, Gainesville, FL, 32607, United States, praveen@ufl.edu

350C Behavioral Analytics Invited: Social Media Analytics Invited Session Chair: Tauhid Zaman, MIT, Cambridge, MA, 02114, United States, zlisto@gmail.com 1 - Building a Location-based Set of Social Media Users Tauhid Zaman, MIT, 6 Whittier Place, Apt 8M, Boston, MA, 02114, United States, zlisto@gmail.com, Christopher Marks We introduce a method for producing a set of social media users from a specified location. Building on existing methods for identifying user locations, network search, and community detection, we propose an expand-classify methodology which recursively collects users based on their connections, and then classifies the users according to a factor graph model. This factor graph model accounts for the implications of both observed user profile features and social network connections in inferring location. Using geo-located Twitter data for evaluation, we find that our method typically outperforms Twitter’s native user search in producing a location-based set of users. 2 - Determining Group Membership using Social Media Activity Patterns Matthew Robert Webb, US.Army, 361 Beacon Street, Apt.1, Boston, MA, 02116, United States, mrwebb@mit.edu We propose a novel classification algorithm to determine a user’s group membership based on the time series of his or her social media posts and the user’s location. We do this by comparing the user’s pattern of activity to that of a known group. We apply our method to the class of Muslim users whose prayer activity is determined by the position of the Sun in relation to the horizon at a location. By assuming Muslim users do not utilize social media during these prayers, we are able to infer the religious affiliation of unknown user accounts. 3 - Predicting Performance under Stressful Conditions using Galvanic Skin Response Tauhid Zaman, PhD, MIT Sloan, Cambridge, MA, United States, zlisto@mit.edu, Carter Mundell, Juan Pablo Vielma We show that the biological signal known as galvanic skin response (GSR) allows one to predict a subject’s performance under stress. We conduct an experiment where subjects answer arithmetic questions under different stress conditions while having their galvanic skin response monitored. Using only the GSR measured under low stress conditions, we are able to predict which subjects will perform well under high stress conditions, achieving an area under the curve of 0.76. Without using GSR, we do no better than random guessing. 4 - Modeling Dynamic user Engagement for Online Gamers David Scott Hunter, MIT, dshunter@mit.edu We propose a new model for user engagement based on online gameplay data. Our model clusters users into varying groups, where each group corresponds to differing gameplay behaviors. Using these clusters, we are able to construct an engagement score that allows us to accurately predict whether or not a user will buy the next version of the game. Invited: Auctions Invited Session Chair: Brian Baisa, Amherst College, Amherst, MA, 01002, United States, bbaisa@amherst.edu 1 - Post Efficient Auctions and English Auctions for Bidders with Non-Quasilinear Preferences Brian Baisa, Amherst College, Converse Hall, Office 211, Amherst, MA, 01002, United States, bbaisa@amherst.edu We study efficient auction design for a single indivisible object when bidders have interdependent values and non-quasilinear preferences. When the object being sold is a good, there is an efficient ex post implementable mechanism if there is an efficient ex post implementable mechanism in a corresponding quasilinear setting. This result extends results on efficient ex post equilibria of English auctions with quasilinearity to our non-quasilinear setting. However, in procurement settings there is no mechanism that has an ex post efficient equilibrium if the level of interdependence between bidders is sufficiently strong. SB28 350D Mechanism Design and Applications

44