INFORMS 2021 Program Book

INFORMS Anaheim 2021

SB31

3 - Intertemporal Pricing via Nonparametric Estimation: Integrating Reference Effects and Consumer Heterogeneity Hansheng Jiang, University of California, Berkeley, Albany, CA, 94706-2651, United States, Junyu Cao, Zuo-Jun Max Shen We consider intertemporal pricing in the presence of reference effects and consumer heterogeneity. Our research question encompasses how to estimate heterogeneous consumer reference effects from data and how to efficiently compute the optimal pricing policy. We propose a demand model that allows arbitrary joint distributions of valuations, responsiveness to prices, and responsiveness to reference prices among consumers. To learn consumer heterogeneity from transaction data, we use a nonparametric estimation method. We investigate the structure of optimal pricing policies and prove the sub- optimality of constant pricing policies even when all consumers are loss-averse according to the classical definition. We validate our model using real data from JD.com, a large E-commerce retailer and find empirical evidence of consumer heterogeneity. 4 - Advertisement Policies with Consumer Privacy Concerns Shouqiang Wang, The University of Texas at Dallas, Naveen Jindal School Of Man., Richardson, TX, 75080-3021, United States, Can Kucukgul, Ozalp Ozer The hallmark feature of digital advertisement platforms is their capability of keeping track of users’ online browsing activities and using this information to personalize advertisements. Various regulations are established to grant users of these platforms the right to privacy, i.e., they can choose whether to share their personal data with the platforms for advertising purposes. Using an information design framework, we study how an online advertisement platform should design its advertisement policy under such regulatory provisions. We show that when the platform and users’ incentives are sufficiently aligned, it is optimal for the platform to adopt a personalized advertisement policy. Perhaps surprisingly, when the incentives are sufficiently aligned, we find that the right to privacy may in fact reduce the overall user surplus. SB33 CC Room 209A In Person: Revenue Management in Online Matching Platforms General Session Chair: Siddhartha Banerjee, Cornell University, Ithaca, NY, 14853- 3801, United States 1 - Dynamic Matchmaking on Gaming Platforms Judy Gan, Columbia University, New York, NY, 10027-6945, United States, Yash Kanoria, Will Ma We consider a dynamic matching model for gaming platforms. Players arrive stochastically with a skill attribute, the ELO rating. The distribution of ELO is known but the individual’s rating is only observed upon arrival. Matching two players with different skills incurs a match cost. The goal is to minimize a weighted combination of waiting cost and match cost in the system. We investigate a popular heuristic in industry, the Bubble algorithm. The algorithm places arriving players on the ELO line with a growing bubble around them. When two bubbles touch, the two players get matched. We show that, with the optimal bubble expansion rate, the Bubble algorithm achieves a constant factor ratio against the offline optimal cost when the match cost (resp. waiting cost) is a power of ELO difference (resp. waiting time). We use data from a gaming start-up to validate our approach. 2 - A Fluid Approximation for a Matching Network with Reneging Angelos Aveklouris, The University of Chicago Booth School of Business, Chicago, IL, 60637-1610, United States, Amy R. Ward Motivated by service platforms, we study a two-sided network where heterogeneous demand and supply arrive randomly over time to get matched and may be lost if forced to wait too long for a match. We develop a fluid model that approximates the evolution of the stochastic model and it is shown that a fluid- scaled state descriptor approaches a solution of the fluid model. Moreover, we study the long-run behavior of a fluid model solution by characterizing the invariant points and showing that a fluid model solution approaches an invariant point in the steady-state. Further, the fluid and steady-state limits can be interchanged. When matches have different values and letting demand and supply wait be costly, we propose a matching policy and show that it is asymptotically optimal on the fluid scale.

SB31 CC Room 208A

In Person: Interpretable and Fair Machine Learning/Spatial & Temporal Analytics and Applications II General Session Chair: Na Zou, Texas A&M University, College Station, TX, 77845, United States Chair: Jian Liu, University of Arizona, Tucson, AZ, 85719-0505, United States 1 - Presenter Na Zou, Texas A&M University, College Station, TX, 77845, United States Attribution methods have been developed to understand the decision making process of machine learning models, especially deep neural networks, by assigning importance scores to individual features. Existing attribution methods often built upon empirical intuitions and heuristics. There still lacks a general and theoretical framework that not only can unify these attribution methods, but also theoretically reveal their rationales, fidelity, and limitations. To bridge the gap, in this paper, we propose a Taylor attribution framework and reformulate seven mainstream attribution methods into the framework. Based on reformulations, we analyze the attribution methods in terms of rationale, fidelity, and limitation with three principles. Finally, we empirically validate the Taylor reformulations on benchmarking real-world datasets. 2 - Spatial-temporal Pose Estimation in Robotic Assembly Process Yinwei Zhang, University of Arizona, Tucson, AZ, 85705-4772, United States, Jian Liu Robotic assembly process usually relies on accurate 6-DoF (Degree of Freedom) pose estimation of the workpieces from streaming images collected by moving robots. The performance of conventional pose estimation methods based on feature extraction may not be accurate in uncertain implementation environment, such as an assembly station with noisy lighting condition. In order to achieve robust estimation, a new method is proposed based on a deep learning model. The proposed method incorporates 3-Dimension CAD models and takes the temporal correlation of consecutive images into consideration. The simulation results show that the proposed method gives improved pose estimation accuracy in the noisy manufacturing environment. SB32 CC Room 208B In Person: New Topics in Revenue Management/First-Price Auctions in Online Advertising Markets General Session Chair: Can Kucukgul, The University of Texas at Dallas, Richardson, TX, 75080, United States 1 - Learning in Repeated First-price Auctions Zhengyuan Zhou, Stern School of Business, New York University, New York, NY, 10012, United States First-price auctions have very recently swept the online advertising industry, replacing second-price auctions as the predominant auction mechanism on many platforms. This shift has brought forth important challenges for a bidder: how should one bid in a first-price auction, where unlike in second-price auctions, it is no longer optimal to bid one’s private value truthfully and hard to know the others’ bidding behaviors? We discuss our recent online learning based approaches to this problem, with the goal of maximizing the cumulative surplus (valuation minus the bid) over time. 2 - Contextual First-Price Auctions with Budgets Rachitesh Kumar, Columbia University, New York, NY, United States, Santiago Balseiro, Christian Kroer The internet advertising market is a multi-billion dollar industry, in which advertisers buy thousands of ad placements every day by repeatedly participating in auctions. In recent years, the industry has shifted to first-price auctions as the preferred paradigm for selling ad slots. A ubiquitous feature of these auctions is the presence of campaign budgets, which specify the maximum amount the advertisers are willing to pay over a specified time period. We present a new contextual model to study the equilibrium bidding strategies in first-price auctions for advertisers who satisfy budget constraints on average. We show the existence of a natural value-pacing-based Bayes-Nash equilibrium under mild assumptions, and study its structural properties. Furthermore, we generalize the existence result to standard auctions and prove their revenue equivalence.

10