2016 INFORMS Annual Meeting Program

TA43

INFORMS Nashville – 2016

TA43 208A-MCC Data-Driven Decision Making Sponsored: Decision Analysis Sponsored Session

2 - An Expected Utility Approach For The Mean-variance Portfolio Problem Felipe Macedo de Morais Pinto, Universidade Federal de Pernambuco, Caixa Postal 7471, Recife, 50630971, Brazil, felipe_mmp94@hotmail.com, Adiel T de Almeida Filho This paper presents an expected utility approach for decision makers with exponential utility behavior as an alternative to the mean-variance approach when considering a financial portfolio. The DA framework is used for modeling the classical Markowitz’s portfolio decision problem incorporating a Bayesian perspective, which allows to include aspects such as the evaluation of macroeconomic environment and minimizing the Bayes Risk. A numerical application is presented based on financial data for an investment decision evaluating a portfolio of DOW 30, FTSE 100 and NASDAQ 100. 3 - A Bayesian Approach For Consumer Credit Debt Collections Process Adiel T de Almeida Filho, Assistant Professor, Universidade Federal de Pernambuco, Caixa Postal 7471, Recife, 50630971, Brazil, adieltaf@cdsid.org.br, Mee Chi So, Christophe Mues, Lyn C Thomas After a borrower defaults on their repayment obligations, collectors of unsecured consumer credit debt have a number of actions they can take to secure some repayment of the debt. The operations management challenge in this setting is to decide which of these actions to take, how long to take them, and in what sequence to take them. In this paper, a Bayesian Markov Decision Process (MDP) model is used to find an optimal policy of what action to undertake in the next period given the current information on the individual debtor’s repayment performance thus far. 4 - An Analytic Method For Investment Analysis In Mulichannel Retailing Somayeh Yasamin Salmani, Drexel University, 2007 Chestnut Street, Apt D2, Philadelphia, PA, 19103, United States, ss3858@drexel.edu, Fariborz Partovi We propose a two-stage stylized model to help firms in making a major strategic decision in distribution channels investment. Our study is motivated by firms that provide multiple channels for customers. We develop an analytic model using customer input and operating costs for specific channel structures to find optimal investing allocation across different distribution channels. TA45 209A-MCC Efficient Learning in Stochastic Optimization Sponsored: Simulation Sponsored Session Chair: Ilya O. Ryzhov, University of Maryland, 4322 Van Munching Hall, College Park, MD, 20742-1815, United States, iryzhov@rhsmith.umd.edu 1 - Continuous Learning For Contextual Bandits With Nonstationary Rewards John G Turner, University of California Irvine, Irvine, CA, United States, john.turner@uci.edu, Amelia C Regan, Tianbing Xu, Yaming Yu We study how best to match ads to viewers using high-dimensional contextual features (demographic, browsing behavior) to predict click-through probability. Using Thompson Sampling in a Bayesian framework, our model learns the importance of contextual features while adapting/forgetting over time, capturing changing individuals’ tastes and shifts in the viewing population’s composition. 2 - Bayesian Bandits For Sequential Clinical Trials Of Multiple Technologies Ozge Yapar, University of Pennsylvania, Philadelphia, PA, United States, yapar@wharton.upenn.edu, Stephen E Chick, Noah Gans We extend recent work on fully sequential trials for health technologies that explore the potential benefits of linking Phase III trials with health technology assessments for market access. We take a bandit perspective that uses Bayesian learning about multiple health technologies.

Chair: Hiba Baroud, Assistant Professor, Vanderbilt University, 2301 Vanderbilt Place, PMB# 351831, Nashville, TN, 37212, United States, hiba.baroud@vanderbilt.edu 1 - Combining Data And Weakly Informative Priors To Make Better Decisions Faster Adam Jason Fleischhacker, University of Delaware, ajf@udel.edu In tackling decision problems, a decision maker must choose how to represent uncertainty using techniques that may be classified on a spectrum; on one end you have fully specified distributions which make strong assumptions, and on the other end, completely non-parametric and robust approaches which minimize assumptions. In this work, we develop and use an analytically tractable model of uncertainty that can model mild assumptions and which can more rapidly extract value from data than non-parametric approaches. 2 - Projection Of Drought Risk For Thermoelectric Power Plants Using Downscaled Climate Scenarios Royce Francis, Assistant Professor, George Washington University, Washington, DC, 20052, United States, seed@email.gwu.edu Many climate researchers have studied a number of climate forcing scenarios to determine how the coupled oceanic-atmospheric systems will respond. At the same time, these responses will be part of a complex feedback loop with infrastructure systems. Thus, it is important to help infrastructure decision makers incorporate climate scenarios into risk and reliability assessments. This presentation demonstrates a Copula Bayesian Network for projecting thermoelectric power plant drought risk over CMIP5 downscaled climate scenarios. 3 - Data-driven Decision Analysis Model For Planning And Management Of Multiple Purpose Reservoir Cascade Systems Thushara De Silva, Vanderbilt University, 400 24th Avenue South, The objective of this study is to develop a decision analysis model for the planning and management of water resources that maximizes multiple objectives such as economic viability, environmental sustainability, and social development. The model is deployed to the Mahaweli water resources development which is the largest multipurpose project of Sri Lanka. A multicriteria decision analysis model is considered and several data sources are used to assess the multiple attributes in the model. The utility function incorporates the preferences of multiple decision makers to assess the weights on the attributes. 4 - Using Data In Decision Making: Big Data, Little Data, No Data Hiba Baroud, Vanderbilt University, hiba.baroud@vanderbilt.edu The role of data analytics in decision making has evolved as the volume of data changed and the tools and technologies to handle such data improved. Are decision makers overwhelmed with data or do they still lack the amount of data they need to improve their decision models? This work is a review of the current state of the art of the use of data-driven tools in decision analysis techniques in practice and theory. The objective is to identify gaps between data and decisions while highlighting opportunities and challenges in research. TA44 208B-MCC Investment Analysis and Financial Applications Sponsored: Decision Analysis Sponsored Session Chair: Manel Baucells, University of Virginia Darden School of Business, 100 Darden Blvd, Charlottesville, VA, 22903, United States, baucellsm@darden.virginia.edu 1 - Net Present Value Analysis And Individual Utility Manel Baucells, University of Virginia Darden School of Business, baucellsm@darden.virginia.edu, Sam Bodily Standard investment analysis employs expected Net Present Value discounting at a risk-adjusted market return. Such prescription takes the viewpoint of the capital market, but neglects the risk aversion of the project owner or the individual investor. We develop an approach that is consistent with expected utility, and requires the integration of project and market returns. The approach recommends the use of the certainty equivalent discount rate, which depends on both the market and the risk aversion of the individual. We explore conditions in which market returns can be omitted from the analysis; or in which our approach particularizes into the standard analysis. 267 Jacobs Hall, Nashville, TN, 37212, United States, thushara.k.de.silva@vanderbilt.edu, George Hornberger, Hiba Baroud

248