Informs Annual Meeting 2017

TE03C

INFORMS Houston – 2017

TE03

Carlo simulation. Of 3 variance reduction methods — change of measure, acceptance-rejection, and conditional Monte Carlo, numerical experiments suggest that conditional Monte Carlo should be preferred.

310C Competition in Multi-Echelon Systems

Invited: Tutorial Invited Session 1 - Competition in Multi-Echelon Systems Awi Federgruen, Columbia University, New York, NY, United States, af7@gsb.columbia.edu Around the start of this new millennium, scholars in the operations management/operations research field started to make important contributions to the study of price competition models. In this tutorial, we review these contributions, and partition them into five broad areas. Most of the tutorial is devoted to the last and most recent category: price competition models for multi- echelon supply chains with an arbitrary number of competing firms and products at each echelon. TE03A Grand Ballroom A Joint Session APS/SIM:Simulation and Applied Probability Sponsored: Applied Probability Sponsored Session Chair: Shane Henderson, Cornell University, Ithaca, NY, 14853, United States, sgh9@cornell.edu 1 - High Volume Traffic Modeling: Simulation and Statistics Zeyu Zheng, Stanford University, 94 Thoburn Ct. 110#, Stanford, CA, 94305, United States, zyzheng@stanford.edu, Peter W. Glynn Many service system applications require the ability to adequately model and statistically characterize high volume traffic. Most prior work with focus at the inter-arrival time scale may fail to identify non-stationarities, correlations and non-Poisson behavior that may only manifest over longer time scales. Existing limit theory shows that system performance is primarily affected by arrival behavior over time scales that are at the order of service time. We develop a “top- down” modeling toolbox to capture the statistical characteristics of incoming traffic that have a primary impact on performance. This talk describes our modeling perspectives and tools. This work is joint with Peter Glynn. 2 - Robust Multivariate Extreme Event Analysis Xinyu Zhang, Columbia University, New York, NY, United States, xz2691@columbia.edu, Henry Lam We discuss an approach to estimate extremal quantities from data under multivariate settings, based on estimated marginal information such as density estimates and tail probability masses, and conjectured geometric constraints such as monotonicity and convexity that aim to reduce estimation conservativeness. The approach utilizes optimization over probability distributions to obtain extremal bounds with statistical confidence guarantees. We discuss our solution techniques that generalize some classical mixture representations for convex sets of probability measures, and the implications of our results in rare-event simulation. 3 - Combined Acceptance-rejection and Importance Sampling Methodologies for Perfect Sampling from Gibbs Point Processes Sarat B. Moka, Visiting Fellow, Tata Institute of Fundamental Research, HB Raod, Colaba, Mumbai, 400005, India, sarathmoka@gmail.com, Sandeep Juneja, Michel Mandjes We consider generating perfect samples from spatial Gibbs processes and associated unbiased estimation problem. Traditionally, the former problem is addressed using coupling from the past based methods. We focus on acceptance- rejection with and without importance sampling for generating perfect samples. We illustrate the efficacy of proposed methods relative to existing ones in a simpler setting of hard sphere models that we analyze in asymptotic regime where the number of spheres generated increases to infinity while the sphere radius decreases to zero at varying rates. Our analysis identifies the large deviations rates of no overlap probability of spheres when centers form a Poisson process. 4 - Estimating the Probability that a Function Observed with (simulation) Noise is Convex Shane Henderson, Cornell University, School of ORIE 230 Rhodes Hall, Cornell University, Ithaca, NY, 14853, United States, sgh9@cornell.edu, Nanjing Jian Consider a real-valued function that can only be observed with stochastic simulation noise at a finite set of design points. We wish to determine whether there exists a convex function that goes through the true function values at the design points. We develop an asymptotically consistent Bayesian sequential sampling procedure that estimates the posterior probability of this being true. In each iteration, the posterior probability of convexity is estimated using Monte

TE03B Grand Ballroom B

Choice Models and Assortments Sponsored: Revenue Management & Pricing Sponsored Session Chair: Sumit Kunnumkal, Indian School of Business, Hyderabad, 500032, India, sumit_kunnumkal@isb.edu 1 - Learning to Rank an Assortment of Products Kris Johnson Ferreira, Harvard Business School, Morgan Hall 492, Boston, MA, 02163, United States, kferreira@hbs.edu, Shreyas Sekar A large product variety makes it impractical for consumers to browse all products before purchasing, making product ranking an important decision. Best practice is to rank products in decreasing order of popularity. However, its optimality assumes that consumers know their expected utility for all products a priori. When this assumption does not hold, consumers must form an expectation of their utility of unseen products. We identify the optimal product ranking in light of this and propose and test an online algorithm that learns consumer preferences and converges to the optimal ranking. 2 - Relating the Fixed Cost and Space Constrained Assortment Problems Jacob Feldman, Olin Business School, United States, jbfeldman@wustl.edu, Alice J.Paul In the assortment optimization problem, a retailer seeks the revenue maximizing set of products to offer to each arriving customer. In effort to make the problem more realistic, there have been a number of extensions to this classic version. For example, the space constrained assortment problem places a limit on the total space consumed by the set of offered products. The fixed cost assortment problem associates a stocking cost with each offered product. Combinatorially, the two problems are drastically different, however, in both problems, there is a “penalty” of sorts for offering a product. The goal of this research is to link the two problems. 3 - Assortment Optimization with Product Costs and Constraints We consider the assortment optimization problem under the MNL model with product fixed costs and constraints. We propose a new method to obtain an upper bound on the optimal expected profit. We show that our method is tractable and has provable performance guarantees for some common types of assortment constraints. TE03C Grand Ballroom C Empirical Operations Management Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Eduard Calvo, ecalvo@iese.edu Co-Chair: Weiming Zhu, IESE Business School, Avenida Pearson 21, Barcelona, 08034, Spain, zhuwm923@gmail.com 1 - The Incentive Game under Target Effects: Evidence from the Ride Sharing Market Liu Ming, University of Maryland, 9530 Baltimore Avenue, Apt 603B, College Park, MD, 20740, United States, mldshg@gmail.com, Xirong Chen, Zheng Li, Weiming Zhu Ride sharing platforms usually provide monetary incentives to keep registered drivers active. To theoretically and empirically analyze the incentive game under target effect, we model platform’s bonus scheme and driver’s working hours at equilibrium. We derive theoretically how driver’s target effects shape their willingness to drive. We then structurally probe the existence of target effect and derive the optimal bonus strategy. Sumit Kunnumkal, Queen’s University, 465 Goodes Hall, Kingston, ON, K7L2G8, Canada, sk162@queensu.ca, Victor Martinez de Albeniz

409