2016 INFORMS Annual Meeting Program

MC45

INFORMS Nashville – 2016

MC44 208B-MCC Panel: Howard Raiffa: Celebration of His Life and Achievement Sponsored: Decision Analysis Sponsored Session Moderator: Jeffrey Keisler, University of Massachusetts - Boston, 100 Morrissey Boulevard, Boston, MA, 02125, United States, jeff.keisler@umb.edu 1 - Howard Raiffa: Celebration Of His Life And Achievement Jeffrey Keisler, University of Massachusetts, Boston, MA, United States, jeff.keisler@umb.edu Colleagues of Howard Raiffa will discuss aspects of his life, contributions and legacy. 2 - Panelist David Bell, Harvard Business School, Morgan Hall 171, Boston, MA, 02163, United States, dbell@hbs.edu 3 - Panelist Ralph Keeney, Duke University, Fuqua School of Business, San Francisco, CA, 94111, United States, keeneyr@aol.com 4 - Panelist Detlof Von Winterfeldt, University of Southern California, USC, Los angeles, CA, 90089, United States, winterfe@usc.edu MC45 209A-MCC Optimization via Simulation under Input Uncertainty Sponsored: Simulation Sponsored Session Chair: Eunhye Song, Northwestern University, 2145 Sheridan Road, Evanston, IL, 60208, United States, eunhyesong2016@u.northwestern.edu 1 - Data-driven Construction Of Uncertainty Sets For Joint Chance- constrained Programs Jeff Hong, City University of Hong Kong, jeffhong@cityu.edu.hk Henry Lam, Zhiyuan Huang We study the use of robust optimization (RO) in approximating joint chance- constrained programs (CCP), in situations where only limited data, or Monte Carlo samples, are available in inferring the underlying probability distributions. We introduce a procedure to construct uncertainty set in the RO problem that translates into provable statistical guarantees for the joint CCP. This procedure relies on learning the high probability region of the data and controlling the region’s size via a reformulation as quantile estimation. We show some encouraging numerical results. 2 - Asymptotics Of Risk Formulations For Simulation Optimization Under Input Uncertainty Di Wu, Georgia Institute of Technology, Atlanta, GA, United States, dwu80@gatech.edu, Enlu Zhou Input distributions are the distributions (of stochastic uncertainty) that drive a simulation process. Since input distributions are usually estimated from finite data, optimizing the model may yield solutions that perform poorly under the true input distributions. To hedge against the risk of input uncertainty, we minimize the risk measures of mean output with respect to the unknown parameters’ posterior distribution. We establish the consistency and the asymptotic normality of risk formulations, and show that when the input data has a small size, the risk formulations are essentially seeking a tradeoff between average performance and the risk of actual performance. 3 - An Optimization-based Approach To Input Uncertainty Via The Empirical Likelihood Henry Lam, University of Michigan, khlam@umich.edu Huajie Qian We study a simulation-optimization-based approach in constructing statistically accurate confidence bounds for stochastic simulation under nonparametric input uncertainty. This approach utilizes the empirical likelihood method that converts the computation of confidence bounds into a pair of optimizations over the uncertain input distributions, with a suitable weighted-average divergence constraint calibrated with a chi-square quantile. We present the theory giving rise to the constraint and the calibration, and compare the performance of our optimization algorithm with existing standard methods such as the bootstrap.

2 - When Do Bidders Anticipate Regret During Auctions? Empirical Evidence From Ebay

A. Serdar Simsek, The University of Texas at Dallas, Richardson, TX, United States, Serdar.Simsek@utdallas.edu, Ozalp Ozer, Meisam Hejazi Nia We developed a structural model that accounts for bidders’ learning and their anticipation of winner and loser regrets in auctions. Using a large data set from eBay, we quantify in which product categories bidders anticipate regret and show how our results can be used to increase eBay’s revenue significantly 3 - Optimal Pricing In Social Networks Under Asymmetric Information Yang Zhang, Tsinghua University, yangzhanguser@mail.tsinghua.edu.cn, Ying-ju Chen We study the optimal pricing of products / services in social networks, where customers are strategic and their consumptions exhibit local externality. Our model concerns the information asymmetry — Consumers know about their local network characteristics while the selling firm has only knowledge of global network. The network model we employ embeds random network and scale free network as special cases. We characterize the optimal menu for the firm, the induced sales, and their properties with regard to the network structure. MC43 208A-MCC Decision Analysis Arcade I Sponsored: Decision Analysis Sponsored Session Chair: Joshua Woodruff, University of Texas, Austin, TX, United States, joshua.woodruff@utexas.edu 1 - Assessment Of Drug Development Options: If, When And How Much? New approaches in clinical trial design and changing regulatory and payer environment make it a challenging task to compare different drug development paths. We will describe a modeling approach to assess different options in time, risk and value dimensions. This will cover uncertainties around success measures, clinical trial timelines and market share expectations. Ideas of eliciting information from subject matter experts as well as combining expert opinion with statistical estimates will be shared. The talk will also reflect on our experience of utilizing different metrics and visualizations to communicate with stakeholders. 2 - Optimal Discretization For Decision Analysis Joshua Woodruff, University of Texas, joshua.woodruff@utexas.edu Optimal discretization is a new method to discretize uncertainties. By using optimization techniques to discretize uncertainties, it is possible to create robust discretizations that are more accurate. Our method produces more accurate project certain equivalents which will improve decision quality. We use non- linear optimization to select and assign probabilities to candidate percentiles for each model uncertainty. With optimal discretization we found we can use the model information and potential distributions of the uncertainties to find discretizations that yield certain equivalent errors that can be orders of magnitude better than other discretization methods we tested. 3 - Network Interdiction In Competitive Market Entry And Product Design Benjamin Harris, Northeastern University, 360 Huntington Avenue, Boston, MA, 02115, United States, harris.be@husky.neu.edu, Sagar S Kamarthi Optimal strategy for product design in the Internet of Things (IoT) must consider input beyond that of stakeholders and customers and include the highly connected infrastructure on which the product will be released. Firms involved in the IoT need to develop strategy and risk mitigation techniques and remain competitive. This research will enable firms to identify optimal strategies under current requirement and market conditions, as well as analyze changes in strategy if a requirements space is changed. As a result, product designers and firm leadership can anticipate and respond to market and industrial changes with increased fidelity and predictive power through network model insights. Ozgur Ozkan, Decision Science Director, Astrazeneca Pharmaceuticals, Gaithersburg, MD, United States, Ozgur.Ozkan@astrazeneca.com

197