Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC63

n TC63 West Bldg 103B

n TC64 West Bldg 104A Data Analytics for Behavioral Operations and Business Decisions Sponsored: Data Mining Sponsored Session Chair: Mohsen Ahmadian, University of Massachusetts Boston, Malden, MA, 02148, United States 1 - The Impact of Sellers’ Reputation on Customers’ Purchasing Intention Toward Remanufactured Products Zhaoji Yu, Shenyang University of Technology, Shenyang, China Zhaoji Yu, California State University, Stanislaus, Turlock, CA, United States, Yingzi Zhao, Xun Xu Apart from elevating public awareness of sustainable consumption, remanufacturing offers an opportunity for products to create a “secondary valueö. Since remanufacturing is a manufacturing process, the ultimate goal is to promote the product sales. Laying emphases on fixed-price trading environment, this paper studies the influence of remanufactured product seller’s reputation on the customer’s willingness to purchase a remanufactured product. The results of the study confirm that consumers are more willing to buy products from manufacturers than are remanufactured by sellers. Interestingly, we also find that the longer the seller registered, the lower its consumer willingness to buy. 2 - Forecasting and Inventory Control with Compound Poisson Demand Using Periodic Demand Data Dennis Prak, University of Groningen, Nettelbosje 2, Groningen, 9747 AE, Netherlands, Ruud Teunter, Mohamed Z. Babai, Aris Syntetos, John E. Boylan We propose a consistent, closed-form method-of-moments estimator for compound Poisson demand parameters that dominates in terms of estimation accuracy and achieved service level. It estimates the arrival rate from the fraction of periods without demand, and combines this with the average of period demand to estimate the mean demand size. It outperforms standard method-of-moments for intermittent demand patterns and performs similarly to maximum likelihood. Period forecasting techniques as implemented in software are not suitable for inventory control based on compound Poisson demand, and doing so leads to severe biases and too high inventories. 3 - Predicting Order Variability in Inventory Decisions: A Model of Forecast Anchoring Dayoung Kim, Assistant Professor, California State University Fullerton, 800 N. State College Blvd, SGMH 5373, Fullerton, CA, 92831, United States, Andrew M. Davis, Li Chen We take a unique perspective to the Newsvendor problem, by developing a simple theoretical model that can accurately predict order variability (of human subjects), in addition to average orders. Indeed, a number of behavioral models which can predict average newsvendor orders (i.e. the pull-to-center effect) have been proposed, but they cannot explain and predict order variability observed in data. Therefore, from a managerial standpoint, our model allows upstream suppliers the ability to form a complete forecast distribution of downstream orders, leading to better inventory decisions, planning, and increased profitability. 4 - Multi-product Demand Prediction and Price Optimizatation Yasaman Shahi, PhD Student, Georgia Institute of Technology, 755 Ferst Dr, Atlanta, GA, 30318, United States, Kris Johnson Ferreira, Gurhan Kok Multi-product demand prediction models typically lack either flexibility or interpretability. We propose a neural network architecture that balances these two and show how it can be used in price optimization. We partner with a consumer electronics retailer to estimate our demand model and implement our price optimization approach. 5 - Rival Chasing Behavior in Simultaneous Competitions Mohsen Ahmadian, PhD Student, University of Massachusetts Boston, 100 William T. Morrissey Blvd., College of Management, Boston, MA, 02125, United States, Roger H. Blake, Ehsan Elahi This research uses the results of laboratory experiments in which subjects playing the role of suppliers competing for the business of a buyer. These results show significant differences with predictions from theory. We find the observed behaviors can be explained by a phenomenon that we name “rival-chasingö. Rival-chasing is individuals’ tendency to change their decisions toward their competitors’ decisions. In symmetric competitions, this tendency leads to a faster convergence of the decisions. However, in asymmetric competitions, rival-chasing behavior results in a closer gap between observed decisions compared with the gap between the equilibrium points predicted by the theory.

Joint Session DM/Practice Curated: Data Science for Electrical Markets Sponsored: Data Mining Sponsored Session Chair: Mehrdad Pirnia, University of Waterloo, 200 University Ave West, Waterloo, ON, N2L3G1, Canada 1 - Internet of Things for Power Consumption Rachneet Kaur, Student, University of Illinois-Urbana Champaign, Department of Industrial Engineering, TB 21, Champaign, IL, 61820, United States, Richard Sowers, Kevin Thompson The objective of our research is to use historical pricing data from a U.S. energy supplier to schedule policies which can be implemented in a smart home which allows for a major appliance to be turned on so as to minimize cost. The policy is back tested, showing significant savings. We design a dishwasher that can be activated via voice command through the use of an app, wireless speaker (Amazon Echo Dot), and an AI voice assistant (Amazon Alexa). More importantly, we want to predict the future price of power in order to activate the dishwasher at an optimal time to save the consumer on the operation cost of the dishwasher. 2 - Data-driven Strategies for Trading Renewable Energy Production Miguel A. Muñoz, PhD. Student, University of Malaga, Edificio I+D Ada Byron, Malaga, Spain, Juan Miguel Morales, Salvador Pineda In this talk we first introduce the problem of selling renewable energy in day- ahead electricity markets with a dual-price balancing settlement as a newsvendor problem. We then analyze different strategies to include information of auxiliary variables that may have predictive power on the renewable power production. Different data-driven optimization techniques are used to determine the optimal amount of energy to be sold in the market taking into account the information of those variables. The performance of these techniques is evaluated in a realistic case study in which we consider a single renewable power producer trading the whole wind power production of a relatively small country. 3 - A Bilevel Optimization Model for Estimating Utilities of Price-responsive Electricity Consumers Arnab Roy, PhD Candidate, University of Louisville, 3177 South 3rd Street, Louisville, KY, 40214, United States, Lihui Bai, Chaosheng Dong, Bo Zeng A bilevel optimization model is presented to estimate parameters of the utility function for price-responsive electricity consumers. The lower level minimizes the sum of electricity cost and inconvenience cost, due to consumers’ curtailment of load in a demand response event. Real-world data from a field demonstration project is used in a joint predictive and prescriptive model (Bertsimas and Kallus, 2015). The upper level minimizes the difference between the total consumption determined by the lower level problem and the consumption measurements from the collected data. 4 - Chance Constrained Optimization to Model Micro-grids with Renewables and Storage Capacities Mehrdad Pirnia, University of Waterloo, 200 University Ave West, Waterloo, ON, N2L3G1, Canada, Hassan Shavandi, Alberto J. Lamadrid, John David Fuller In this presentation, we develop a social welfare maximization model with storage capacities (batteries), considering wind uncertainties. The uncertainties associated with wind generation is modeled using chance constraints. Furthermore, an equivalent deterministic mixed integer programming model is used to find the solution of the chance-constrained model. We use these models to analytically investigate the impact of storage capacities on the electricity prices, and the optimal number of charge/discharge for batteries. We also present the electricity price differences between the stochastic and deterministic models.

330