Volume 54, Issue 5

In This Issue

Evaluating Academic Programs
The public media such as Business Week and the Wall Street Journal periodically produce rankings of many types of academic programs. These rankings are widely distributed and perceived to influence the quality of applicants, prestige of schools, and financial donations to programs. Yet most of the evaluations on which the rankings are based have significant flaws and biases in both substance and process. Among these are the arbitrary selection and weighting of criteria for evaluation and the unappraised self-reported information from academic institutions. In “Evaluating Academic Programs: With Applications to U.S. Graduate Decision Science Programs,” R. L. Keeney, K. E. See, and D. von Winterfeldt develop a methodology to evaluate academic programs that avoid such flaws and biases. The methodology is used to separately evaluate prescriptive and descriptive U.S. graduate decision science programs. The paper illustrates procedures to evaluate both disciplinary and interdisciplinary academic programs and suggests how it might be applied to evaluate operations research programs.

A Practical Method for Improving Anti-Cancer Drug Treatment Schedules
Cancer is an increasingly prevalent disease. As novel approaches, such as gene therapy, still face significant hurdles before they can become established therapeutic strategies, current control of cancer largely depends on drug therapy. But drug therapy is a complex problem, involving many interactive non-linear processes, which operate on different organization levels of the biological system. Therefore, for assessing the effect of each potential treatment on the patient prognosis, clinicians must be provided with methods for formal and systematic analysis of the intricate interactions between the drug and the hierarchy of bodily processes. In “Optimizing Chemotherapy Scheduling Using Local Search Heuristics,” Z. Agur, R. Hassin, and S. Levy introduce a practical heuristic method for predicting the most efficacious treatment schedules, according to the criteria set by the clinician. The novelty in the method lies in the link it makes between operations research methodology and biomathematical approaches to modeling quantitatively retrievable biological processes. The suggested optimal schedule for the chemotherapeutic agent, taxol, can be experimentally tested, thus paving the way for clinical validation of the treatment optimization method put forward there.

A New Approach to Piecewise Linear Optimization
Separable piecewise linear functions arise in optimization models where they are used to approximate nonlinear functions. In many applications, such as supply chain design, the functions are not convex and auxiliary binary variables are required which leads to a mixed-integer programming formulation. In “A Branch-and-Cut Algorithm without Binary Variables for Nonconvex Piecewise Linear Optimization,” A. B. Keha, I.R. de Farias Jr, and G. L. Nemhauser give a branch-and-cut algorithm to solve this class of problems. However, instead of using a MIP model, they formulate the problem with only continuous variables and apply specialized branching and cuts to handle the nonconvexities. They give computational results that demonstrate that this approach is superior to the MIP approach.

Dynamic Dual-Channel Optimization
Manufacturers in some industries, such as electronic components, industrial equipment, and steel, have access to electronic marketplaces, where potential customers post items they wish to buy their chosen prices. These electronic orders can be used to complement the manufacturers’ contracts with their long-term customers. In “Revenue Management of a Make-to-Stock Queue,” R. Caldentey and L. M. Wein consider a queuing model for a risk-averse manufacturer who chooses a long-term contract price, observes the deterministic demand for its product at this price, and then solves a make-to-stock queuing control problem for accepting or rejecting electronic orders that arrive according to a renewal process with spot-market prices modeled as an exponential reflected random walk that is correlated with the demand process. Under heavy traffic conditions, they incorporate two aspects of mathematical finance –geometric Brownian motion and portfolio optimization – into a traditional operations management model, and then use analytical approximations to derive a simple accept/reject policy that depends on both the spot price and the current finished goods inventory level.

Dynamic Control of Make-to Order Manufacturing System
While traditionally manufacturers keep inventory to fulfill demand, many are adapting the make-to-order strategy so as to better satisfy the varying needs of their customers. Although make-to-order strategy allows manufacturers to offer greater variety of products, it also results in increased lead times. Therefore, a key challenge in operating make-to-order manufacturing systems is to optimize system performance while keeping lead times at acceptable levels. In “Dynamic Control of a Multiclass Queue with Thin Arrival Streams,” B. Ata studies the problem of dynamically controlling a make-to-order manufacturing facility with due-date lead-times, and develops an explicit nested threshold policy for controlling such a system.

Dynamic Cross-Selling on the Internet: A New Source of Revenues
Cross-selling on the Internet is a recent phenomenon that is quickly gaining popularity. For example, an attempt to buy most books on Amazon.com will generate a suggestion to buy a package of two or more books. Moreover, many companies offer a discount on cross-sold products. The challenge with cross-selling products on the Internet is that it has to be performed dynamically in response to every customer’s purchase attempt. At the same time, the choice of products to cross-sell must be made in real time by the software on the basis of information about product inventories and customer preferences. In “Revenue Management Through Dynamic Cross-Selling in E-Commerce Retailing,” S. Netessine, S. Savin and W. Xiao propose a novel modeling framework for the dynamic cross-selling problem in which combinatorial optimization (package selection) and stochastic dynamic programming (package pricing) are applied. They obtain insightful structural properties of the problem and compare several efficient heuristics that allow solving an otherwise formidable problem in real time. The authors argue that the dynamic cross-selling offers yet untapped source of additional revenues to the companies.

Dynamic Pricing in Make-to-Order Systems
How should a manufacturer integrate dynamic pricing decisions with those of production planning and control? What is the optimal product mix for the manufacturer and in what sequence should it be produced? How should product prices be adjusted in response to increasing order backlogs? What information is needed in integrating pricing and production decisions? Revenue management has transformed the transportation and hospitality sectors over the past couple of decades, and is now becoming increasingly important in many other industries. “Revenue Management for a Multi-Class Single-Server Queue via a Fluid Model Analysis,” by C. Maglaras, studies how to apply these successful demand management ideas in manufacturing. It does that by developing a mathematical model that jointly selects dynamic pricing and production strategies to optimally trade-off revenues with inventory costs in order to maximize profitability. The results show that production decisions should be made in order to minimize the total holding costs irrespective of the firm's pricing strategy, while dynamic pricing adjustments should be made in response to an aggregate measure of congestion given by the outstanding workload in the system in a way that accounts for product substitution effects. Implementing such a strategy requires that the pricing function is informed of the aggregate workload of the firm and the cost parameters associated with the production process.

Scalable Simulations for a Supply Network
The prototype of a re-entrant factory is the semiconductor fab, which in 2005 is a multibillion dollar investment. In order to improve their performance or to test new control or decision algorithms, detailed simulation tools at the level of discrete event simulations have successfully been developed. However, these simulations are extremely time consuming. Recently, the emphasis has shifted from improving the production process in an individual factory to improving the overall performance of the entire supply chain – suppliers, production factories and distribution networks. The goal is to study time dependent “what if” scenarios for the whole supply network, e.g., “what happens if we ramp up product X in factory Y and ramp it down in factory Z?” Discrete event simulation takes far too long to be a useful decision tool for these questions. In “A Continuum Model for a Re-Entrant Factory,” Armbruster, et al, present a heuristic attempt to model the core piece of such a supply network – the re-entrant fab – via a model similar to a fluid flow through a pipe. The simulation time of this model is independent of the number of machines and the number of products run in the factory. The paper shows that the model compares well with discrete event simulations, discusses control issues arising from such a model and proposes its embedding into a supply chain model.

Time Windows in Production Systems
In industries where traceability is a major concern, where products are perishable, or where stock facilities have limited capacity, preparation of customer demand should not start before a given release date, and should be ready before its due date. Such siturations lead N. Brahimi, S. Dauzère-Pérès, and N. Najid to the introduction of the notions of “production time windows”. In “Capacitated Multi-item Lost-Sizing Problems with Time Windows,” the authors propose mathematical formulations to model a generic problem that represents the above mentioned problems, among others. Their model considers the case where several products compete for the utilization of a single resource with limited capacity. They solve the problem using a Lagrangian relaxation-based heuristic, which gives very efficient results.

The Spiral-Down Effect in Revenue Management
Airlines use revenue management models to control the availability of different products (e.g., ticket classes). Some widely-used models assume that the random demand for each product is exogenous, that is, that the probability distribution of this demand is not affected by the airline’s booking control. However, customers choose among available products, and thus an airline’s booking control usually affects the probability distribution of the demand for each product. In “Models of the Spiral-Down Effect in Revenue Management”, W. L. Cooper, T. Homem-de-Mello, and A.J. Kleywegt study the dynamics of revenue management forecasting and optimization processes that are controlled by a revenue manager who uses a model based on such an exogenous demand assumption. They show that, in such a situation, attempts by the revenue manager to refine parameter estimates may actually make revenue management performance systematically deteriorate in the long run, and in some cases can lead to a “spiral down” of both protection levels for higher priced products and total revenues. These results point out the importance of developing and implementing accurate models of consumer behavior, as well as the need to identify models and estimation procedures that are robust against departures from their assumptions.

Reservation Policies for Improved Inventory Control in Divergent Systems
In many industries customers place orders with the expectation of delivery at a later date. This deviates from the traditional assumption in inventory modeling, where the customer requests immediate delivery of every item demanded. The presence of advance-order information about future deliveries raises interesting questions regarding how to use this temporal information to improve the replenishments and allocation of inventories throughout the supply chain. In “Controlling Inventories in Divergent Supply Chains with Advance-Order Information,” J. Marklund analyzes these questions in the settings of a one-warehouse multiple-retailer system operating in a continuous review environment. Assuming that the advance-order information is made available throughout the supply chain, the focus is on reservation policies that allow for differentiating the warehouse service to different retailers. A crucial question is when should an item be reserved for a specific retailer or customer? Exact and approximate cost evaluation methods for three warehouse reservation policies are presented. A numerical study illustrates their performance and provides insights as to the value of using reservation times as a means for warehouse inventory allocation.