Informs Annual Meeting 2017

SD03A

INFORMS Houston – 2017

2 - Foundations of Discounted Satiation Manel Baucells, PhD, Darden School of Business, University of Virginia, 100 Darden Blvd, Charlottesville, VA, United States, BaucellsM@darden.virginia.edu We provide axiomatic foundations for a representation of preferences that relaxes the time separability axiom of discounted utility. Our goal is to account for diminishing marginal utility in a dynamic setting where recent past consumption decreases the marginal utility of current consumption. We clarify the relationships between satiation, durability and time-separable preference, and offer an analytical characterization of the optimal consumption profile that maximizes satiation preferences. 3 - Range Dependent Utility for Risk and Time Manuel Baucells, Darden School of Business, University of Virginian, Charlottesville, VA, United States, BaucellsM@darden.virginia.edu We propose Frame Dependent Utility, a behavioral model to evaluate risky prospects, possibly with temporal delay. Frame dependent utility is able to explain the most robust anomalies (four-fold pattern, common ratio and common consequence, hyperbolic discounting, and the magnitude effect), as well as the preference reversal phenomenon which has eluded many descriptive theories of decision making. The model is simpler to use than cumulative prospect theory because it does not require sorting outcomes from highest to lowest. 4 - Revisiting Ellsberg and Machina’s Paradoxes: A Two-Stage Evaluation Model under Ambiguity Ying He, University of Southern Denmark, Campusvej 55, Odense, M, 5230, Denmark, yinghe@sam.sdu.dk We revisit the Ellsberg paradox by showing that the preferences satisfy the independence axiom. Such an observation motivates us to develop a two-stage evaluation model under ambiguity. Events in state space are classified into ambiguous and risky events. We define a special type of risky event as hedge. In our model, a conditional subjective expected utility (SEU) on each hedge is nested into an SEU to represent preferences over acts. We show that our two-stage model can not only accommodate preferences in different Ellsberg’s paradoxes but also the preferences in paradoxes in Machina (2009) and Machina (2014) that challenge the validity of many existing models for decision making under ambiguity. 310B Data-Driven Bayesian Modeling Sponsored: Decision Analysis Sponsored Session Chair: Debarun Bhattacharjya, IBM, T J Watson Research, Yorktown Heights, NY, 12, United States, debarun@us.ibm.com Co-Chair: Hiba Baroud, Vanderbilt University, Nashville, TN, 37240, United States, hiba.baroud@vanderbilt.edu 1 - Design Optimization for Resilience for Risk-averse Firms Ramin Giahi, Iowa State University, Ames, IA, United States, rgiahi@iastate.edu, Cameron MacKenzie Engineered systems can face various adverse conditions during their lifetime. Designers should try to design systems that are resilient to these conditions. This research quantifies the resilience of an engineered system under time-dependent adverse conditions by simulating the degradation and recovery of the system’s components. The firm incorporates the future conditions under which the system operates into a risk-averse utility function or via a value-at-risk measure. Since the optimization model requires a complex simulation to evaluate the objective function, we use Bayesian Optimization, in conjunction with the ranking and selection method, to optimize the design of the system. 2 - Bayesian Decision-making of Smart Systems under Uncertainty Saideep Nannapaneni, Vanderbilt University, Nashville, TN, United States, saideep.nannapaneni@vanderbilt.edu, Abhishek Dubey, Sankaran Mahadevan This talk discusses a model-based framework for the quantification of multiple uncertainty sources affecting the performance of a smart system. With strong feedback coupling between the subsystems of a smart system (plant, cyber, actuation, sensors), the uncertainty in the system output amplifies over time. The coupling occurs at two levels: (1) coupling between individual subsystems, and (2) coupling between computational nodes. The coupled smart system is decoupled and considered as a feed-forward system over time and modeled using a two-level Dynamic Bayesian Network; one at each level of coupling. The proposed methodology is demonstrated for the design of a smart infrastructure system. SD02

3 - Hierarchical Bayesian Modeling of Infrastructure Network Recovery Mackenzie G. Whitman, Graduate Research Assistant, Vanderbilt University, Nashville, TN, 37203, United States, mackenzie.g.whitman@vanderbilt.edu, Hiba Baroud Risk managers are interested in measuring and improving the ability of critical infrastructure systems to withstand and recover from disruptions which constitutes to the system’s resilience. This work utilizes hierarchical Bayesian methods to assess the recovery rate of network components using multiple data sources to estimate recovery parameters, offering probabilistic instead of deterministic outcomes to better account for uncertainty. We apply this method to three interdependent infrastructure systems that represent the water, electric, and gas distribution networks for Shelby County, Tennessee. 4 - Event Sequence Models for Scenario Recommendations Debarun Bhattacharjya, IBM.TJ.Watson Research Center, 501 East 78th Street, Apt. 2A, New York, NY, 10075, United States, debarunb@us.ibm.com, Nicholas Mattei Predicting the future is a ubiquitous human endeavor, practiced from antiquity to modern professions including business, finance and intelligence analysis. Our intent is to build systems that can support long-term predictions using knowledge generalized from the past and learned from a variety of sources including historical event datasets. In this talk, I will present some ongoing work on models for predicting future event sequences. The emphasis will primarily be on probabilistic models for prediction, but I will also discuss other approaches such as prediction by analogy. 310C Metabolic Networks and Modern Research Problems in Operations Research Invited: Tutorial Invited Session Chair: Jiming Peng, University of Houston, Houston, TX, 77204, United States, jopeng@Central.uh.edu Chair: Rajan Batta, University at Buffalo (SUNY), 410 Bell Hall, Buffalo, NY, 14260, United States, batta@buffalo.edu 1 - Metabolic Networks and Modern Research Problems in Operations Research J. Paul Brooks, Virginia Commonwealth University, Dept of Stat. Science and OR, P.O. Box 843083, Richmond, VA, 23284, United States, jpbrooks@vcu.edu, Allen Holder Biochemical reactions that sustain life are modeled as a system of differential equations, forming a metabolic network. These nonlinear systems curtail into linear, algebraic systems under asymptotic assumptions, which thus define the feasible flux states of a cell in equilibrium. An optimization problem is then imposed to further identify metabolic conditions in an optimal growth situation. The result is a linear programming (LP) technique called Flux Balance Analysis (FBA), and the primary goal of this tutorial is to illustrate several modern research questions about FBA that are compliant with operations research. We pose several open avenues for research that touch LP, mixed-integer programming (MIP), nonlinear programming, combinatorial optimization, optimization under uncertainty, and robust optimization. SD03A Grand Ballroom A Stochastic Dynamics on Networks and Graphs Sponsored: Applied Probability Sponsored Session Chair: Kavita Ramanan, Brown University, Providence, RI, 02912, United States, kavita_ramanan@brown.edu 1 - High Frequency Trading and Limit Order Book Dynamics: from Point Processes to Stochastic Pdes Rama Cont, Imperial College London, London, SW7 2AZ, United Kingdom, Rama.Cont@imperial.ac.uk The advent of high frequency electronic trading has lead to a heterogeneous market landscape where algorithms at different trading frequencies interact through the limit order book, whose evolution summarizes the dynamics of supply, demand and price. We show that the dynamics of the limit order book may be described by a stochastic partial differential equation (SPDE), obtained as scaling limit of a queueing model with high and low frequency traders. The dynamics of bid/ask prices is endogenous and derived from the dynamics of order flow. We exhibit an analytically tractable class of solution and show that they reproduce empirical features of limit order book profiles and price dynamics. SD03

99