Informs Annual Meeting 2017

WB18

INFORMS Houston – 2017

WB18

3 - Pricing and Availability Strategies with Strategic Customer Behavior and Scarcity Effects Hanqing Liu, City University of Hong Kong, Hong Kong, work.lhq@gmail.com, Peng Hu, Stephen Shum We study the dynamic pricing and availability decisions of a firm that repeatedly introduces new generations of a product over time. Customers are strategic and are affected by two types of scarcity effects: direct scarcity effect and relative scarcity effect. Firm-induced scarcity is never optimal in the absence of scarcity effects but it can be optimal when customers are affected by scarcity effects. We study the long-run optimal policy of the firm and characterize when firm-induced scarcity is optimal. Our results show that the firm’s long-run optimal policy may be a constant availability policy or a varying availability policy, depending on certain parameters. 4 - To Buy or Not to Buy? Purchasing Decision and Spillover Effect of the Sharing Economy for the Manufacturing Industry Jie Yang, Phd Candidate, Tianjin University, Nankai District, 92 Weijin Road, Tianjin, 300072, China, nemoyj1989@126.com, Daozhi Zhao We first study how the presence of sharing economy affects the consumer’s purchasing decision. We focus on this question, and identify the conditions for consumers to purchase products. In the presence of the conspicuous consumption, we develop a model in which consumers make purchase decisions and then have the option to participate in a sharing economy contingent on a realized usage level. We compare the consumers’ purchase decisions to a situation where ex-post sharing is impossible. Second, we investigate the spillover effect of the sharing economy on the manufacturing industry. 340A Stochastic Models for Shared Transportation Sponsored: Applied Probability Sponsored Session Chair: Siddhartha Banerjee, Cornell University, Ithaca, NY, 14853, United States, sbanerjee@cornell.edu 1 - The Value of State Dependent Control in Ride-sharing Systems Pengyu Qian, Columbia Business School, Columbia Business School, c/o PhD Office, New York, NY, 10027, United States, PQian20@gsb.columbia.edu, Yash Kanoria, Siddhartha Banerjee We consider the design of state-dependent control policies for a closed queueing network model of ride-sharing systems, where the platform can choose which car to match to any incoming passenger. We propose a gradient-based control policy associated with a novel Lyapunov function, and show that this is rate-function optimal, i.e., under mild conditions, it minimizes the asymptotic rate of demand- dropping as the number of vehicles scales to infinity. We further show how this large-deviations approach can be used to study the trade-off between estimation, computation and performance. 2 - Minimizing Multimodular Functions and Allocating Capacity in Bike-sharing Systems Daniel Freund, 1, 109 Lake St, Ithaca, NY, 14850, United States, danielrfreund1990@gmail.com, Shane Henderson, David B. Shmoys Bike-share systems (BSS) allow users to rent and return a bike at any station within the system. Whereas the question of how to mitigate the effects of resulting imbalances using rebalancing has received much attention, our work considers the more strategic question of how to allocate dock capacity within such systems. Extending known models from the rebalancing literature and connecting it to the literature on discrete convexity, we present a provably polynomial algorithm that finds the optimal allocation of docks within such systems. Our work has yielded improvements for the BSS in major American cities like Boston, Chicago, and NYC, where strategically reallocating docks has improved service. 3 - Online Matching in a Ride Sharing Platform We study the online problem of matching passengers with each other for sharing rides. Our setting is motivated by real-world ride-sharing platforms like Lyft. In this setting, a passenger can request a ride at any time, and the task of the matching algorithm is to find another ride to match with it. A ride can get matched only to another ride that arrived close in time, and there is a bounded time interval to match a ride since its arrival. The efficiency of a match between two given rides defines the weight of the match. The goal of the matching algorithm is to maximize the total weight of the matching constructed over a given time interval. We present algorithms and bounds for this problem. WB17 Chinmoy Dutta, Lyft, San Francisco, CA, United States, cdutta@lyft.com, Adam Greenhall, Keshav Puranmalka, Chris Sholley

340B Applied Probability Contributed Session Chair: Suneel Babu Chatla, National Tsing Hua University, Hsinchu, Taiwan, suneel.chatla@iss.nthu.edu.tw 1 - Revisiting the Two-sample Z-test from the Perspective of Almost Stochastic Dominance Chin Hon Tan, National University of Singapore, 1 Engineering Drive 2, Department of Industrial and Systems Engineer, Singapore, 117576, Singapore, isetch@nus.edu.sg The two-sample z-test is often used to compare the mean of two populations. However, such tests are not sufficient for studying preferences between two uncertain prospects. In particular, the z-test only accounts for sampling errors and neglects the risk preferences of the decision maker. In practice, the preferences of the decision maker is unclear. Here, we revisit the two-sample z-test in light of recent advances in the almost stochastic dominance literature which expound on the relationship between preferences and distribution moments. 2 - Video Streaming in Distributed Erasure-coded Storage Systems Vaneet Aggarwal, Purdue University, 315 N. Grant St., West Lafayette, IN, 47906, United States, vaneet@purdue.edu, Abubakr Al-Abassi This work considers video streaming over distributed systems where the video segments are encoded using an erasure code for better reliability thus being the first work to our best knowledge that considers video streaming over erasure- coded distributed cloud systems. The download time of each coded chunk of each video segment is characterized and ordered statistics over the choice of the erasure-coded chunks is used to obtain the playback time of different video segments. Using the playback times, bounds on the moment generating function of the stall duration is used to bound the mean stall duration. This metric represents an important quality of experience (QoE) measure for the end users. 3 - Additive Model for Unbalanced Longitudinal Data via Smooth Backfitting Suneel Babu Chatla, PhD Candidate, National Tsing Hua University, Hsinchu, 30013, Taiwan, suneel.chatla@iss.nthu.edu.tw, Li-Shan Huang We propose a new method for estimating an additive model for the unbalanced longitudinal data using smooth backfitting. Our method uses the concept of global smoothing matrix and it enables estimating the regression function at the data points in addition to the grid points. We use profile least square technique to estimate the correlation structure over repeated measurements. We prove that the proposed estimator is asymptotically efficient compared to the estimator obtained using working independence assumption. A Monte Carlo simulation study was conducted in order to evaluate the finite sample performance of the estimators. A real data application is also provided. 4 - Predictive Control and Fault Detection of Natural Gas Networks Kevin Melendez, PhD Student, University of South Florida, 4411 Shady Terrace Lane, Apt 303, Tampa, FL, 33613, United States, kmelendez7@mail.usf.edu, Tapas K. Das, Changhyun Kwon The objective of this study is to generate a comprehensive methodology to optimally design and operate a natural gas network. A Mixed Integer Nonlinear Programming model is proposed to locate the pipelines and compression stations while minimizing the total cost over a planning horizon. A Bayesian approach is used for demand forecasting and a model predictive control strategy is introduced for maintaining the network on the optimal operation point. Finally, multivariate statistics methods are used for early fault detection. The proposed methodology aims to help the engineers of a natural gas company with the decision-making and the online control and atypical operating condition identification. 5 - Confidence Sets for Split Points in Model Based Regression Trees Suneel Babu Chatla, National Tsing Hua University, Hsinchu, 30013, Taiwan, suneel.chatla@iss.nthu.edu.tw, Galit Shmueli Model-based (MOB) trees are popular for their ability to incorporate parametric models into classification and regression trees. The existing MOB uses an exhaustive search to identify the split points and it is computationally very intensive. In addition, there is no way to assess the variability of the identified split points, which is crucial for the statistical generalizability of the results. To solve this problem, we propose a new split point estimation method which uses the change point framework. We consider two popular bootstrap methods to estimate the confidence sets: conditional and smooth. From a simulation study, we find that the smooth bootstrap provides much narrower confidence sets.

479