Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SB19

2 - Dynamic Procurement of New Products with Covariate Information- The Residual Tree Method Gah-Yi Ban, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, Jeremie Gallien, Adam J. Mersereau We study the practice-motivated problem of dynamically procuring a new, short life-cycle product under demand uncertainty. The firm does not know the demand for the new product but has data on similar products sold in the past, including demand histories and covariate information such as product characteristics. We propose a new, combined forecasting and optimization algorithm called the Residual Tree method, and analyze its performance via epi- convergence theory and computations using data from Zara. Our method generalizes the classical Scenario Tree method by using covariates to link historical data on similar products to construct demand forecasts for the new product. 3 - Designing Response Supply Chain Against Bioattacks Yun Zhang, MIT, Cambridge, MA, United States, Nikolaos Trichakis, David Simchi-Levi This work studies the problem of prepositioning medical inventory against bioattacks. We model it as a general two-stage inventory optimization problem on a network, and provide tractable and optimal solution methods. We collect data from publicly available sources and perform a thorough case study against anthrax attacks. 4 - Data-driven Order Assignment for Last Mile Delivery Sheng Liu, University of California, Berkeley, 1731 Spruce St Unit B, Berkeley, CA, 94709, United States, Long He, Zuo-Jun Max Shen Motivated by a food delivery service provider, we discuss a data-driven framework to model the delivery performance and optimize the order assignment decision using real delivery data. Leveraging on classical results in routing literature and machine learning approaches, we propose a prediction model for the delivery tour distance. We then propose a sample average approximation model and a distributionally robust model for order assignment. A branch-and- price algorithm is developed to solve both models efficiently. In the numerical study, we show the benefits of data-driven order assignment models integrated with delivery tour prediction, compared to classical vehicle routing models. n SB18 North Bldg 128A New Topics in Supply Chain Management Sponsored: Manufacturing & Service Oper Mgmt/Supply Chain Sponsored Session Chair: Serguei Netessine, The Wharton School, Philadelphia, PA, 19104, United States 1 - Supply Chain and Antitrust Governance: Supply Chain and Antitrust Governance: Can Contractual Agreements Reinforce the “Illinois Brick”? Nitish Jain, London Business School, Sussex Place, Regent’s Park, London, NW1 4SA, United Kingdom, Sameer Hasija, Serguei Netessine, Serguei Netessine In 1977, the U.S. Supreme Court issued a key ruling in Illinois Brick Co v Illinois case that debarred indirect purchasers to sue for recovery of antitrust damages. This ruling, however, can be exploited by firms to form collusion; muting the effectiveness of private enforcers in fighting antitrust violations. In this paper, using a supply chain lens, we characterize the conditions that facilitate or deter collusion formation. Our findings provide actionable insights to public enforcers in active case selection and investigation of antitrust violations. 2 - Do Flexibility and Chaining Really Help? An Empirical Analysis of Automotive Plant Networks Vivek Choudhary, INSEAD Business School, 1 Ayer Rajah Avenue, Singapore, Singapore, Sameer Hasija, Serguei Netessine We study production networks of automotive assembly plants to shed new light on the impact of flexibility on performance. We reconcile the extant empirical and modeling literature by testing existing flexibility indices. We introduce new flexibility indices to assess the relationship between flexibility and productivity to explain the trend of real-life networksWe empirically show that intermediate levels of flexibility are optimal due to trade-off between better flexibility and productivity, however, chaining may not be the optimal configuration if changeover losses of flexible plants are accounted for. 3 - Worker Poaching in a Supply Chain: Enemy from Within? Gad Allon, University of Pennsylvania, 3730 Walnut Street, Philadelphia, PA, 19104, United States, Evan Barlow, Achal Bassamboo Poaching workers has become a universal practice. We explore worker poaching between firms linked in a supply chain. We show that the classical intuition from labor economics is insufficient in explaining poaching between supply chain partners. We also show how and under what conditions worker poaching can

actually improve supply chain performance. Finally, we show how the equilibrium identity of the supply chain bottleneck depends on the interaction between hiring, poaching, and productivity. 4 - Sustaining Forests and Smallholders by Eliminating Payment Delay in a Commodity Supply Chain Dan Andrei Iancu, Associate Professor, Stanford University, 655 Knight Way, Stanford, CA, 94107, United States, Joann de Zegher, Erica Plambeck Motivated by the Indonesian palm oil industry, we propose a profitable way in which large buyers can simultaneously improve farmer livelihoods and halt illegal deforestation in their supply chains.Currently, farmers are paid with delays,and buyers avoid sourcing from illegally-deforested land by monitoring individual farmers.Instead, we propose rewarding all farmers in a village by eliminating payment delay if no production occurs on illegally-deforested land in the village.Using field data, dynamic programming and game theory,we show how this village-level incentive improves productivity and profitability for farmers, processors and buyers, and best halts illegal deforestation. n SB19 North Bldg 128B Digital Ad Auctions Sponsored: Revenue Management & Pricing Sponsored Session Chair: Hemant K. Bhargava, University of California, Davis, CA, 95616, United States Co-Chair: Ramnik Arora,Facebook, Menlo Park, CA, 9, United States 1 - No Really, What Does Any of this Mean?: Proxy Metrics in Ad Auctions Dan Chapsky, Facebook, New York, NY, United States Digital ad auctions are rife with data available for interpretation by marketers. In general, some combination of data on clicks, views, conversions is used to interpret which strategies are actually delivering value. Much research has been done to show that the only true method of measuring value is through randomized control trials, but the majority of online ads go through no such experimentation. In this talk we will be discussing exploratory research on which non-experimental measurement methods work and when. 2 - Pacing Mechanisms for Ad Auctions Nicolas Stier-Moses, Facebook, 1 Hacker Way, Menlo Park, CA, 94025, United States, Vincent Conitzer, Christian Kroer, Debmalya Panigrahi, Okke Schrijvers, Eric Sodomka, Chris Wilkens Online ad auction platforms typically rely on pacing mechanisms to ensure that campaign daily budgets are consistent with maximum bids. The goal is maximizing advertisers’ utilities. We model this process and connect it to market equilibria. The goal of this system is being able to forecast market outcomes. 3 - Budget-constrained Ad Selection in Large-scale Advertising Systems Hemant K. Bhargava, University of California, Gh-3108, Graduate School of Management, Davis, CA, 95616, United States, Ramnik Arora Internet advertising systems operate at large scale with tens of thousands of potential ads mapping to an individual or a keyword. Moreover, search results need to be computed in a fraction of a second to provide seamless user experience. We report results of experiments with a hierarchical decomposition approach where candidate ads are sequentially pruned using a finer and more computationally expensive machine learning model and/or higher number of user / ad features. 4 - On Optimal Auctions for Mixing Exclusive and Shared Matching in Platforms Hemant K. Bhargava, University of California, Gh-3108, Graduate School of Management, Davis, CA, 95616, United States, Gergely Csapo, Rudolf Muller We study auctions that can produce either exclusive or shared allocations, and involve two dimensions of information, a bidder’s valuations for exclusive and shared allocation, respectively. We formulate the revenue-maximizing incentive- compatible auction, and demonstrate that our proposed solution, a two-dimensional reserve-price based mechanism (RM), has excellent revenue performance.

37