In This Issue

Abstract

“In this Issue” offers a brief overview of the articles appearing in the current issue of Operations Research , highlighting their applicability to current problems in operations research faced by the entire OR community: practitioners, researchers, educators, and students. Light-Speed Trading The wheels of finance have started to turn faster. Modern equity markets are currently dominated by high-frequency traders that expend significant resources to develop a state-of-the-art trading infrastructure. This infrastructure allows trading with remarkable speed: the latency of an order, or the time it takes for an order to be processed on a modern exchange, has decreased dramatically from minutes to microseconds over the past few decades. In “The Cost of Latency in High-Frequency Trading,” C. Moallemi and M. Sağlam explore underlying drivers behind this phenomenon. They describe an analytical model that allows for the quantitative valuation of latency as a market friction using two well-known market parameters: volatility and the bid-ask spread. Empirical estimation suggests that latency is not only important for high-frequency traders, as thought of by many before, but for all large cost-sensitive investors. Fair Division of a Maximum Flow in a Bipartite Network Fairly dividing a maximum-flow in a bipartite network is a fundamental problem in operations research. In “The Bipartite Rationing Problem,” H. Moulin and J. Sethuraman study this problem from an axiomatic viewpoint, with consistency being the key requirement: the division of the max-flow is unchanged if an edge is dropped from the problem, and the supply and demand of its end points are adjusted down by the flow on this edge. When there is a single supply node, this is the well-known problem of rationing a single resource on which agents have subjective claims. Some well-known consistent fair division methods for this problem are the proportional method, the uniform gains method, and the uniform losses method. The authors extend these methods to the case of a general bipartite network; they show that the proportional method admits a unique consistent extension, whereas the uniform gains and uniform losses methods admit infinitely many. The Analytic Hierarchy Process for Multicriteria Decision Making Decisions involve satisfying many criteria, some in contradiction with each other, and the alterations of choice must be evaluated in terms of these criteria, some or all of which may be intangible. We need a way to measure alternatives on intangible criteria that is in harmony with tangible measures. In science we have mostly measured so the numbers' relative values have meaning. But when we divide two ratio scale numbers to get their relative value we no longer have a ratio scale number but an absolute number that is invariant under the identity transformation. By using judgment to estimate such absolute numbers, we could uniformize the measurements and trade them off using the weights or priorities of the criteria. In “The Modern Science of Multicriteria Decision Making and Its Practical Applications: The AHP/ANP Approach,” T. L. Saaty presents the Analytic Hierarchy Process (AHP), which is based on the mathematics of these ideas. Product-Generation Transition and Global Capacity Expansion Decision Analysis Production capacity decisions are often subject to a host of uncertainties related to both the demand for the new product generation and the commercial supply that is released into the marketplace. Supply-constrained environments can arise for biotechnology drugs where difficulties in manufacturing and limited competition are commonplace. Current analytical practice uses the market-share attraction model to forecast sales. This works well when there is enough supply to meet demand. However, this assumption is unrealistic in a supply-constrained environment. In “Product-Generation Transition Decision Making for Bayer's Hemophilia Drugs: Global Capacity Expansion Under Uncertainty with Supply-Demand Imbalances,” J. S. Stonebraker applies decision analysis to global production capacity expansion under uncertainty for Bayer's proposed new biotechnology drug to treat hemophilia A. This decision analysis resulted in an improved approach to Bayer's decision that more realistically addressed potential regional supply-demand imbalances when supply is a limiting factor and there are substantial uncertainties about both demand and supply. Formulation for Dynamic Traffic Assignment Problems Dynamic traffic assignment (DTA) problems are traditionally formulated as both path-based and link-based models. In “An Intersection-Movement-Based Dynamic User Optimal Route Choice Problem,” J. Long, H.-J. Huang, Z. Gao, and W. Y. Szeto propose a novel variational inequality formulation for the dynamic user optimal (DUO) route choice problem using approach proportion. Approach proportions contain travelers' route information so that the realistic effects of physical queues can be captured in a formulation when a physical-queue traffic flow model is adopted, and so that route enumeration and path-set generation can be avoided in the solution procedure. In addition, the simple structure of the approach proportion representation allows us to decompose the constraint set for solving large-scale DUO route choice problems. The authors show that the proposed DTA model is consistent with some other DTA models, and that the proposed method can outperform the path-based method in terms of efficiency and memory usage. Efficiency Measurement in Multioutput Production Settings In “Opening the ‘Black Box’ of Efficiency Measurement: Input Allocation in Multioutput Settings,” L. Cherchye, B. De Rock, B. Dierynck, F. Roodhooft, and J. Sabbe develop a new data envelopment analysis–based methodology for measuring the efficiency of decision making units (DMUs) in multioutput environments. Their method explicitly includes information about output-specific inputs and joint inputs in the efficiency evaluation. This method contributes to opening the “black box” of efficiency measurement in two ways. First, including information on the input allocation substantially increases the discriminatory power of the efficiency measurement. Second, it allows for decomposing the efficiency value of a DMU into output-specific efficiency values, which facilitates the identification of the outputs the manager should focus on to remedy the observed inefficiency. An application to a unique data set collected from the activity-based costing system of a large service company demonstrates the practical usefulness of the methodology. Tools for Proving Monotone Structural Policies in Dynamic Programming Problems Supermodularity plays an important role in deriving monotone structural properties of optimal solutions in parametric optimization problems. In dynamic settings, lattice structures are often required on the underlying feasible sets so that supermodularity can be preserved under optimization operations. In “Preservation of Supermodularity in Parametric Optimization Problems with Nonlattice Structures,” X. Chen, P. Hu, and S. He identify some nonlattice structures under which supermodularity can be preserved. They establish various properties that unify several results in the literature and provide powerful tools to analyze a variety of dynamic operations models. Efficiently Allocating Simulation Sampling Effort The problem of multiple comparisons with a known standard is common in simulation and arises when we have a collection of stochastic systems and wish to use simulation to determine which of these systems have an expected performance exceeding a threshold. Although it is possible to solve this problem through brute-force sampling methods that sample each system equally often, more sophisticated methods save time by allocating simulation effort adaptively where it will be most beneficial. In “Sequential Bayes-Optimal Policies for Multiple Comparisons with a Known Standard,” J. Xie and P. I. Frazier present an optimal strategy for allocating simulation effort in this problem. Assuming a limited ability to sample, the authors use a Bayesian characterization of the problem and use decomposition techniques to avoid dimensionality in solving the dynamic program that characterizes the Bayes-optimal sampling strategy. They show that this sequential Bayes-optimal sampling policy can improve efficiency in achieving desired comparisons. Adaptively Learning Preferences with Optimization Models Preference learning has been a topic of interest for many years and in many different fields, including operations research, marketing, machine learning, and behavioral economics. In “Learning Preferences Under Noise and Loss Aversion: An Optimization Approach,” D. Bertsimas and A. O'Hair combine the ideas from these different fields into a single methodology to learn preferences and make decisions. In particular, the proposed method uses robust and integer optimization in an adaptive and dynamic way to determine preferences from data that are consistent with human behavior. Additionally, the authors derive uncertainty sets from adaptive questionnaires, use the analytic center of discrete sets (as opposed to polyhedra) to capture errors and inconsistencies, and use efficient methods from computer science for sampling to optimize CVaR. The authors demonstrate the strength of the proposed approach via an online software system. Efficient Real-Time Assignment of Virtual Machines in a Network Cloud Cloud computing poses many challenging resource allocation problems. One of them is how to assign virtual machines (VM) to physical machines (PM) in a large data center in real time, so that the total number of PMs in use is minimized. Various combinations of VMs can be “packed” into a single PM, depending on the amount of resources (such as processing speed, memory, disk storage, etc.) they require. VMs arrive at random times and depart after their work is done. A. L. Stolyar proposes a corresponding stochastic model in “An Infinite Server System with General Packing Constraints.” The author shows that certain versions of a natural greedy assignment algorithm are in fact optimal in the asymptotic sense, as the system scale (the average number of active VMs) becomes large. The algorithms are simple, do not need to know the system parameters, and make decisions based on the current system state only. Flows in Networks Under Component Failure Network flows serve as a versatile mathematical tool in operations research. They model flows of passengers, utilities, goods, rolling stock, data, or finance. In many real-world networks, maintenance, overload, malfunction, disastrous events, or malignant attacks can cause failure of nodes and arcs. In “Robust and Adaptive Network Flows,” D. Bertsimas, E. Nasrabadi, and S. Stiller address this issue from the perspective of robust optimization. They study static and dynamic flows that are robust against the failure of arcs. A maximal robust flow maximizes the remaining flow value after the failure of a worst-case set of arcs. In the less conservative adjustable flow model, the flow can be rerouted on the surviving arcs, while not exceeding the initial flow on any arc. They provide insights into the mathematical structure, complexity, approximability, and computability of both models and their corresponding cut problems. Is It Worth Adding Another Machine? The performance of a scheduling system can be improved by increasing the number of processing machines. In “Parallel Machine Scheduling: Impact of Adding Extra Machines,” K. Rustogi and V. A. Strusevich study the impact that additional machines may have if added to the system consisting of identical parallel machines. They measure such a machine impact by the ratio of the value of an objective function (the maximum or the average completion time) computed with the original number of machines to the one computed with extra machines. Tight bounds on the machine impact factors are provided. In addition, the issues of cost-effectiveness of choosing the number of machines are discussed, provided that each machine incurs a cost and the objective function captures the trade-off between the cost of the used machines and a scheduling objective.