Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TE22

2 - Dynamic Assortment Optimization With Inventory Cost Venus Lo, Cornell University, 206 Rhodes Hall, Cornell University, Ithaca, NY, 14853, United States, Huseyin Topaloglu We consider a dynamic assortment optimization problem of selecting an assortment of products and their inventory level to offer to a stream of customers. Unlike traditional revenue management problem where the seller controls the products available to each arriving customer, the seller only makes initial inventory decisions. There is a cost of stocking the inventory, with no salvage value. The assortment available to each customer depends on the initial stock and the purchases of preceding customers, who are willing to substitute. We present a dynamic program for deterministic customer arrivals with nested preferences and develop related heuristics for the case of stochastic customer arrivals. 3 - Inventory Balancing With Online Learning Xinshang Wang, Columbia University, New York, NY, United States, Wang Chi Cheung, Will Ma, David Simchi-Levi We study an online resource allocation problem. A manager needs to allocate limited resources to a heterogeneous pool of customers arriving in real time, while maximizing a certain notion of cumulative reward. The manager can observe a list of feature values of each arriving customer, which allow the manager to learn personalized customer behavior and customize allocation decisions. The challenge is to allocate resources effectively, in the absence of any demand forecast model and under uncertain customer behavior. We propose online allocation algorithms with provably tight performance guarantee. Our technique combines online learning theories with competitive analysis. n TE20 North Bldg 129A Using Pricing to Manage On-line Marketplaces Sponsored: Revenue Management & Pricing Sponsored Session Chair: Robert L. Phillips, Uber, San Francisco, CA, 94101, United States 1 - Surge Moves Drivers: Evidence From a Natural Experiment at Uber Alice Lu, Uber, San Francisco, CA, United States, Peter Frazier, Oren Kislev Ridesharing platforms use dynamic pricing (so-called “surge pricing”) to raise prices when demand from riders outstrips drivers’ availability. This is intended to reduce rider demand and attract drivers. While short-run price increases clearly influence riders, anecdotal evidence on whether they move drivers is mixed, with experienced drivers counselling against “chasing surge”. Using a natural experiment created by a surge pricing service outage affecting a portion of Uber’s driver-partners, we show that visibility of the surge heatmap increases drivers’ earnings, attracts drivers toward areas with higher surge prices, and explains 10%-60% of drivers’ self-positioning decisions. 2 - An Empirical Analysis of Price Formation, Utilization, and Value Generation in Ride Sharing Services Tunay Tunca, Unversity of Maryland, 4361 Van Munching Hall, Robert H. Smith School of Business, College Park, MD, 20742, United States, Weiming Zhu, Liu Ming, Yi Xu Using data obtained from a leading ride-sharing company, we construct and estimate a discrete choice based model of price, demand and supply formation in ride-sharing services based on operational characteristics such as driver utilization and demand intensity. Further, we conduct counterfactual analysis to examine efficiency and welfare implications of regulation and pricing strategies such as surge pricing. n TE21 North Bldg 129B Joint Session RMP/MSOM: Topics in Revenue Management and Learning Sponsored: Revenue Management & Pricing Sponsored Session Chair: Arnoud den Boer, University of Amsterdam, Amsterdam, Netherlands. Anna Saez de Tejada Cuenca, PhD Student, UCLA Anderson School of Management, Los Angeles, CA, United States, Felipe Caro We analyze empirically the drivers of adherence to the prices recommended by a decision support system during the sales season of a fast fashion retailer. We study the effect of two interventions aimed to increase adherence, as well as the drivers 1 - Believing in Analytics: Managers’ Adherence to Price Recommendations From a Decision Support System

of the evolution of adherence over time. One of the firm’s interventions, a change in the way revenue was displayed in the DSS’s interface, increased adherence to its price recommendations. Moreover, managers’ systematic deviations from the DSS’s recommendations are compatible with well-known behavioral biases such as loss aversion and salience of the inventory. Our results provide insights on how to design better DSSs to entice practitioners to use them. 2 - Assortment Rotation and the Value of Concealment Kris Johnson Ferreira, Harvard Business School, Morgan Hall 492, Boston, MA, 02163, United States, Joel Goh We study one reason why frequent assortment rotations can be valuable. Namely, by distributing its seasonal catalog of products over many assortments, the retailer conceals a portion of its product catalog from consumers, injecting uncertainty into the consumer’s relative product valuations. Rationally acting consumers may respond to this uncertainty by purchasing more products, generating additional sales for the retailer. We refer to this phenomenon as the value of concealment. We develop a consumer choice model and stochastic dynamic program to study the value of concealment. 3 - Non-stationary Stochastic Optimization with Applications to Dynamic Pricing Yining Wang, Carnegie Mellon University, 5000 Forbes Avenue, 8203 Gates-Hillman Center, Pittsburgh, PA, 15213, United States, Xi Chen, Yuxiang Wang We consider a non-stationary sequential stochastic optimization problem, in which the underlying cost functions change over time under a variation budget constraint. We propose an $L_{p,q}$-variation functional to quantify the change, which yields less variation for dynamic function sequences whose changes are constrained to short time periods or small subsets of input domain. Under the $L_{p,q}$-variation constraint, we derive both upper and matching lower regret bounds for smooth and strongly convex function sequences, which generalize previous results in Besbes et. al. (15). Our results reveal some surprising phenomena such as the curse of dimensionality of the function domain. 4 - Price Optimization under The Finite-mixture Logit Model Arnoud V. den Boer, University of amsterdam, Science Park 107, Amsterdam, Netherlands, Ruben van de Geer We present a price optimization algorithm under the finite-mixture logit model, a choice model that can capture customer heterogeneity. Our algorithm has theoretical performance guarantees and desirable computational properties. n TE22 North Bldg 130 Application of Revenue Management and Pricing: Interaction of Supply and Demand Sponsored: Revenue Management & Pricing Sponsored Session Chair: Shadi Sharif Azadeh 1 - Robust Demand Estimation under the Multinomial Logit Model Using Sales Transaction Data Sanghoon Cho, University of South Carolina, Columbia, SC, United States, Jongho Im, Pelin Pekgun, Mark Ferguson We propose a statistical procedure that can estimate the effect of product attributes and unobservable lost sales under a choice-based demand model using only historical sales transactions, product availability data, and market share information. Our proposed approach does not require time-homogeneous arrival rates, and allows for varying product attributes and choice sets over time. A set of simulation studies and an application to a real hotel dataset are conducted in comparison to several existing methods. 2 - A Revenue Management Model for on Demand Systems Yousef Maknoon, Delft University of Technology, Delft, Netherlands, Bilge Atasoy, Moshe E. Ben-Akiva, Michel Bierlaire, Shadi Sharif Azadeh This paper presents a revenue management model for mobility on-demand systems. When a customer arrives to the system, the model decides about the mixture of products offered to the customer to maximize the expected revenue. The expected revenue from an offer set is defined by the price and the demand of each offered product. We formulate the problem as a mixed-integer linear model and present a branch and bound algorithm to solve it. Computational results depict the performance of our approach.

387