Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC23

3 - Peer-to-Peer Trading of Usage Quotas Behrooz Pourghannad, University of Minnesota, Minneapolis, MN, United States, Saif Benjaafar, Jian-Ya Ding A growing number of businesses are being built around a model that provides customers access to a product or a service up to a specified amount or allowance. We study the impact of allowing customers to trade among themselves unused portions of their allowances and examine conditions under which peer to peer trading is beneficial to both the firms and consumers. 4 - Customer Learning and Competition for Online Sales of Durable Goods Clark C. Pixton, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States, Clark C. Pixton, Brigham Young University, Provo, UT, United States Motivated by data from a large online retailer, we study online sales of durable goods, focusing on the effects of uncertainty about product quality and learning from customer reviews. We describe the nature of the tradeoff between learning product quality over time and substitution effects between products offered in the same category on the same website. We offer an alternate explanation for market dominance based on transient effects of a learning process which involves decisions under risk, and show that the learning is slower and market dominance more likely in higher price markets. We discuss implications of our model for games between platforms and sellers. Sponsored: Finance Sponsored Session Chair: Tim Siu-Tang Leung, University of Washington, Seattle, WA, 98195-3925, United States 1 - Energy Storage and Trading Daniel Mitchell, PhD, University of Minnesota, Kumar Muthuraman, Stathis Tompaidis, Long Zhao We examine the problem of trading in a commodity that requires storage with transaction costs. We present an algorithm to find the optimal buying, selling and holding regions and consider extensions where prices exhibit seasonality, and when the inventory level can fluctuate randomly, such as a water reservoir that can be filled by rain and depleted through evaporation. 2 - Optimal Dynamic Basis Trading Bahman Angoshtari, PhD, University of Washington, Seattle, WA, United States, Tim Siu-Tang Leung We study the problem of dynamically trading a futures contract and its underlying asset assuming a stochastic basis model. We describe the basis evolution by a scaled Brownian bridge, but also incorporate the possibility of non- convergence at maturity. The optimal trading strategies are determined from a utility maximization problem under hyperbolic absolute risk aversion (HARA) risk preferences. By analyzing the associated Hamilton-Jacobi-Bellman equation, we derive the exact conditions under which the equation admits a solution and solve the utility maximization explicitly. A series of numerical examples are provided to illustrate the optimal strategies and examine the effects of model parameters. 3 - American Options under Periodic Exercise Opportunities Kazutoshi Yamazaki, PhD, Kansai University, Kansai, Japan We study a version of the perpetual American call/put option where exercise opportunities arrive only periodically. Focusing on the exponential Levy models with i.i.d. exponentially-distributed exercise intervals, we show the optimality of a barrier strategy that exercises at the first exercise opportunity at which the asset price is above/below a given barrier. Explicit solutions are obtained for the cases the underlying Levy process has only one-sided jumps. 4 - Dynamic Index Tracking and Exposure Control using Derivatives Tim Siu-Tang Leung, University of Washington, Dept of Applied Mathematics, University of Wa, Lewis Hall #202, Seattle, WA, 98195-3925, United States We develop a methodology for index tracking and risk exposure control using financial derivatives. Under a continuous-time diffusion framework for price evolution, we present a pathwise approach to construct dynamic portfolios of derivatives in order to gain exposure to an index and/or market factors that may be not directly tradable. Among our results, we establish a general tracking condition that relates the portfolio drift to the desired exposure coefficients under any given model. We also derive a slippage process that reveals how the portfolio return deviates from the targeted return. n SC23 North Bldg 131A Stochastic Control Problems in Finance

maximize delivery efficiency, minimize idle shopper hours and minimize orders lost due to lack of supply, all while managing the uncertainty. 2 - Dynamic Pricing with Sales- and Inventory-dependent Demand: The Effectiveness of Certainty-equivalent Approximation Mengzhenyu Zhang, University of Michigan, Ann Arbor, MI, United States, Hyun-Soo Ahn, Christopher Ryan, Joline Uichanco We study a pricing problem where future demand is influenced by past sales and/or product availability as well as price. Under this setting, the price of a product not only determines the revenue and demand in that period but also influences future demands. Hence, the role that a price decision plays is more complicated than most revenue management models where price only affects the sales in that period. We provide asymptotic results that illustrate the effectiveness of certainty-equivalent policies in these settings. We also analytically demonstrate the benefits of dynamic versus static pricing and characterize when it is revenue- optimal to underserve the market by restricting total supply. 3 - Joint Clustering of Retail Products and Customers Andrew Vakhutinsky, Oracle Labs, 35 Network Drive, Sharon, MA, 02067, United States, Daniel Peterson Cluster analysis of products and customers is useful for marketing and product targeting. Using a large dataset of retail transactions for millions of households across one year, we develop a probabilistic clustering of households and products that captures typical purchase behavior. Latent Dirichlet Allocation (LDA) is attractive for this application, because it allows a soft clustering. In this work demonstrate the shortcomings of naive application of LDA, present a few simple solutions that improve the product clusters dramatically, and discuss how metadata about the products and households can be used to further improve the clustering. We also compare results from other approaches. 4 - Score a Retail Store Location Yihui Huang, Tsinghua University, Beijing, China, Chen Wang, Lei Zhao Expected customer visits and sales are two important measures for a retail to decide whether to open a new store at a candidate location. Noticing different shopping behaviors of customers from different types of locations, e.g., shopping malls (S), crowd points (C), offices (O) and resident communities (R), we propose time-domain and frequency-domain models to predict these measures. We develop an expectation-maximization (EM) algorithm to learn the model parameters based on the data of existing stores and public data. n SC22 North Bldg 130 Mechanism Design, Networks, and New Markets Sponsored: Revenue Management & Pricing Sponsored Session Chair: Ozan Candogan, University of Chicago, Chicago, IL, 27708, United States 1 - Optimal Commissions & Subscriptions in Networked Markets Hongfan Chen, University of Chicago, 5050 S. Lake Shore Drive, S3417, Chicago, IL, 60615, United States, Ozan Candogan, Daniela Saban, John R. Birge We consider a platform providing a marketplace to sellers and buyers of different types, whose compatibility is encoded in a bipartite network. With the platform only controlling commissions/subscriptions, we establish that, the platform may need to differentiate the charges to maximize revenues. We show that charging homogeneous commissions/subscriptions or charging to only one side of the market leads to heavy revenue loss in general. Under homogeneous value distributions, we provide a bound on the revenue loss in terms of the network structure. Finally, under some convexity assumptions, we show the revenue- maximizing equilibrium induces at least 2/3 of the maximum social welfare. 2 - Optimal Dynamic Matching under Lost Sales We consider a dynamic matching problem where the agents on one side, say supply side, are long-lived, and those on the other, say demand side, are impatient and will be lost if unmatched upon arrival. The matching reward is the multiplication of quality levels of agents in a match. In a centralized setting, we show that the optimal matching priority is different from assortative mating: it is optimal to prioritize high-quality supply over low-quality one to satisfy high- quality demand; however, for the low-quality demand, low-quality supply has a priority. We show how the decentralized matching process can be perfectly coordinated as the centralized one. Yun Zhou, McMaster University, 1280 Main Street West, Hamilton, ON, L8S 4M4, Canada, Zhiyuan Chen, Ming Hu

69