Seminars

The following speakers will visit Kellogg this Spring as part of the 2013 Kellogg Operations Seminar Series. Please click on a date for more information about a particular talk. The Presentations are at noon in Jacobs, Room 561 unless noted.

View past seminars

"Cohort Turnover and Productivity: The July Phenomenon in Teaching Hospitals"
Talk by Robert S. Huckman, Harvard University and NBER
April 17th, 2013
We consider the impact of cohort turnover - the planned simultaneous exit of experienced employees and a similarly sized entry of new workers - on productivity in the context of teaching hospitals. Specifically, we examine the impact of the annual July turnover of residents in American teaching hospitals on levels of resource utilization and quality in teaching hospitals relative to a control group of non-teaching hospitals. We find that this annual cohort turnover results in increased resource utilization (i.e., higher length of hospital stay) for both minor and major teaching hospitals, and decreased quality (i.e., higher mortality rates) for major teaching hospitals. Specifically in major teaching hospitals, we find evidence of a gradual trend of decreasing performance that begins several months before the actual cohort turnover, which may result from a transition of responsibilities that occurs at major teaching hospitals in anticipation of the upcoming cohort turnover.

(Joint work with Hummy Song and Jason R. Barro)

"Which Suppliers Adhere to Global Labor Standards? Evidence from Codes of Conduct Audits"
Talk by Mike Toffel, Harvard Business School
April 10th, 2013
In response to stakeholder pressures, many transnational businesses have developed codes of conduct and monitoring systems to ensure that working conditions in their supply chains meet global labor standards. Many observers have questioned whether this form of private regulation has any impact on working conditions or is merely a marketing tool to deflect criticism of valuable global brands. We conduct one of the first large-scale comparative studies using codes of conduct audits from one of the world's largest social auditors to determine what combination of government and civil society institutions promote compliance with the global labor standards embodied in these codes. We find that these private transnational governance tools are most effective when they are embedded in states that have made binding domestic and international legal commitments to protect workers' rights and that have high levels of press freedom and nongovernmental organization activity. Taken together, these findings suggest the importance of multiple, robust, overlapping, and reinforcing governance regimes to meaningful transnational regulation.

"Management of Energy Technology for Sustainability: Funding Energy Technology R&D"
Talk by Erin Baker, University of Massachusetts Amherst
March 20th, 2013
Climate change is a major pupblic policy problem. One vexing problem faced by policy makers is how to allocate research budgets across a variety of energy technologies, in order to reduce the future costs of controlling climate change. In this paper we apply a multi-model approach, implementing probabilistic data derived from expert elicitations into a stochastic programming version of a dynamic integrated assessment model, in order to arrive at insights about the optimal government-funded R&D portfolio. We focus on electricity technologies with a significant chance of a breakthrough – solar PV, CCS, and nuclear. We find that the optimal investment is fairly robust to different specifications of climate uncertainty, to different policy environments, and to assumptions about the opportunity cost of investing; and that policy makers would do better to over-invest in R&D rather than under-invest. We show that R&D is even more valuable in “2nd-best” policy environments, when politics, incentives, and uncertainty don’t lead to optimal policies. Finally, we show that R&D can play different roles in different types of policy environments, sometimes leading primarily to cost reduction, other times leading to better environmental outcomes.

"Optimal Design of Social Comparison Effects: Setting Reference Groups and Reference Points"
Talk by Xuanming Su, Wharton
March 13th, 2013
In this paper, we study how social planners should exploit social comparisons to pursue their objectives. We consider two modes of social comparison, referred to as behind-averse and ahead-seeking behaviors, depending on whether individuals experience a utility loss from under-performing or a utility gain from over-performing relative to their peers. Modeling social comparison as a game between players, we find that ahead-seeking behavior leads to output polarization whereas behind-averse behavior leads to output clustering. A social planner can mitigate these effects in two ways, (i) by providing the full reference distribution of outputs instead of an aggregate reference point based on the average output, and (ii) by assigning players into uniform rather than diverse reference groups. Social planners may thus need to tailor the reference structure to the predominant mode of social comparison and their objective. A performance-focused social planner may set the reference structure so as to maximize the output of either the top or the bottom player depending on whether she puts greater marginal weight to larger or smaller outputs. When the social planner also cares about utility, she faces a dilemma because performance-optimization may not be aligned with utility-maximization. Inevitably, the social planner will have to confront equity issues because better performance may not reflect great effort or greater ability.

(Joint work with Guillaume Roels)

"Collaboration in Service Networks: Architectures and Throughput"
Talk by Itai Gurvich, Kellogg
March 6th, 2013
Motivated by the trend towards more collaboration in service work flows, we study processes where some activities require the collaboration of multiple human resources. Collaboration introduces synchronization requirements that are not captured in the conventional procedure to identify bottlenecks and theoretical capacity. We introduce the notions of collaboration architecture and unavoidable idleness. In general, collaboration architectures may feature unavoidable idleness i.e that the theoretical capacity exceeds the maximal achievable throughput or actual capacity. This fundamental tradeoff between collaboration and efficiency does not disappear with scale and has important ramifications to service-system staffing. We identify a special class of collaboration architectures that have no unavoidable idleness. Identifying that the theoretical capacity matches the actual capacity still leaves open the question of how to coordinate/synchronize the resources to achieve that capacity. In collaborative networks, there is a tradeoff between throughput and controllability that cannot be avoided. We study the implications of this tradeoff to the choice of prioritization-and-synchronization policies.

(Joint work with Jan Van Mieghem)

"Structural Estimation of Callers' Delay Sensitivity in Call Center"
Talk by Seyed Emadi, Kellogg
February 27th, 2013
We model callers' decision making processin call centers as an optimal stopping problem. After each period of waiting, a caller decides whether to abandon or to continue to wait. The utility of a caller is modeled as a function of her waiting cost and reward for service. We use a random-coefficients model to capture the heterogeneity of the callers and estimate the cost and reward parameters of the callers using the data of individual calls made to an Israeli call center. We also conduct a series of counterfactual analyses that explore the effects of changes in service discipline on resulting waiting times and abandonment rates. Our analysis reveals that modeling endogenous caller behavior can be important when major changes (such as a change in service discipline) are performed, and that using a model with an exogenously specified abandonment distribution may be misleading.

"Patient Flow Management in Emergency Departments"
Talk by Junfei Huang, NUS, visiting Kellogg
February 13th, 2013
Motivated by its significant impact on quality of care and patient satisfaction, we consider the patient flow management problem in emergency departments (EDs): a choice must be made between triage patients who are yet to be checked vs. those who are in-process (IP). Physicians' capacity is modeled as a queing system with multi-class customers, where some of the classes face deadline constraints on their time-till-first-service, while the other classes feedback through service while incurring congestion costs. We consider two types of such costs: per individual visit to a server or cumulative over all visits. In both cases, we propose and analyze scheduling policies that, asymptotically in conventional heavy-traffic, minimize congestion costs while adhering to all deadline constraints. Via data from the complex ED reality, we use our models to quantify the value of refined individual information, for example whether an ED patient will be admitted to the hospital as opposed to being dicharged. Finally, for our proposed policies, we develop some congestion laws that support forecasting of waiting and sojourn times.

"Integrating Inventory Replenishment and Cash Payment Decisions in Supply Chains"
Talk by Kevin Shang, Duke
October 3rd, 2012
This paper studies the impact of financial integration on a supply chain consisting of a retailer who periodically orders from a supplier in a finite horizon. The retailer and the supplier form a partnership or strategic alliance, aiming to minimize the entire supply chain cost. The retailer pays the supplier for inventory replenishment and decides investment amount in each period. We consider two payment schemes that represent different levels of financial integration. For flexible payment, the retailer is allowed to prepay or delay the payment (with no debt at the end of the horizon); for the strict payment scheme, the retailer pays exactly what she orders. We prove that the optimal joint policy for the flexible payment model has a surprisingly simple structure -- both parties implement a base-stock policy for inventory replenishment; the retailer monitors her cash level and implements a two-threshold policy for investment and a pay-up-to policy for inventory payment. Solving the strict payment model is more involved. We derive a lower bound on the optimal cost by connecting the strict payment model to an assembly system and propose a simple heuristic. Comparing the optimal costs between the flexible and strict payment models yields the true value of payment flexibility. The results of numerical studies suggest that the value of flexible payment can be very significant and that the volatility of material and financial flows may not amplify in the same direction under the flexile payment scheme. Finally, we relate a vertically integrated system to the flexible payment one. Interestingly, we find that the additional benefit achieved by vertical integration may not be significant.

"Double-Counting of Emissions in Carbon-Optimal and Carbon-Neutral Supply Chains"
Talk by Felipe Caro, UCLA
October 24th, 2012
Carbon footprinting is a tool for firms to determine the total greenhouse gas (GHG) emissions associated with their supply chain or with a unit of final product or service. Carbon footprinting typically aims to identify where best to invest in emission reduction efforts, and/or to determine the proportion of total emissions that an individual firm is accountable for, whether financially and/or operationally. A major and under-recognized challenge in determining the appropriate allocation stems from the high degree to which GHG emissions are the result of joint efforts by multiple firms. As more firms make (part of) their supply chains carbon neutral, by choice or by regulation, these allocation questions become more critical.

We introduce a simple but effective model of joint production of GHG emissions in general supply chains, decomposing the total footprint into processes, each of which can be influenced by any combination of firms. A supply chain in which all firms exert their first-best emissions reduction effort levels is "carbon-optimal", while one which offsets all emissions is "carbon-neutral". With this structure, we examine conditions under which a carbon-neutral supply chain will also be carbon-optimal. We find that, in order to induce the carbon-optimal effort levels, the emissions need to be over-allocated, in contrast to the usual focus in the life cycle assessment (LCA) and carbon footprinting literatures on avoiding double-counting. We analyze the problem from the perspective of the social planner as well as that of a "carbon leader", a single firm that offsets all supply chain emissions and that can contract with other firms to encourage them to help reduce emissions. We show that even when the carbon leader can only contract on emissions, she can still induce the same effort levels and profits as when she can contract directly on effort. Our work aims to lay the foundation for a framework to integrate the
economics- and LCA-based perspectives on supply chain carbon footprinting.

(Joint work with Charles Corbett, Tarkan Tan and Rob Zuidwijk)

"Online Stochastic Bin Packing"
Talk by Varun Gupta, Booth
October 31st, 2012
Motivated by the problem of packing Virtual Machines on physical servers in cloud computing, we study the problem of one-dimensional online stochastic bin packing. Items with sizes sampled from an unknown distribution arrive as a stream and must be packed on arrival. The size of an item is known when it arrives and the goal is to minimize the number of non-empty bins. Online stochastic bin packing has been extensively studied in theoretical computer science, combinatorics, and probability literature, and there exist many heuristics. However all such heuristics are either optimal for only certain classes of item size distributions, or rely on learning the unknown distribution. We present a new distribution-agnostic bin packing heuristic that is asymptotically optimal for all distributions, and is extremely simple to implement.

Next, we consider the more general problem of online stochastic bin packing with item departures to which our algorithm extends as-is. We also revisit the popular Best Fit packing heuristic, which has not been studied so far in the setting of item departures.

(Joint work with Ana Radovanovic)

"Pricing Tools and Salespeople -- The Effectiveness of Pricing Decision Support Tools in Business-To-Business Markets"
Talk by Wedad Elmaghraby, Maryland
November 7th, 2012
Wide spread use of information technology to capture and analyze data to support decision making in complex business situations has created a vibrant business segment: There are numerous consulting companies and software providers that specialize in pricing analytics.

Pricing optimization tools have been embraced by business-to-consumer (B2C) companies, such as retailers, airlines, and hotels. The next wave of adaptors of pricing optimization tools are companies in the business-to-business (B2B) arena. Pricing decisions in B2B organizations are not completely automated: Sales people are the ones who quote the prices to the customers and are responsible for "closing the deals" - as sales people typically have a "working relationship" with their customers. Considering the level of involvement of the sales people in the decision making process, and the frequent reluctance of sales people to adapt to change, it is not clear how much benefit a pricing optimization tool can provide in a B2B setting.

We have been given access to sales and pricing data for one of the largest grocery products distributors (GPD) in the United States who used a price recommendation tool over a two year period. We investigate how salespeople use the information provided to them to set the prices; of particular interest to us is how salespeople use price recommendations coming from a decision support tool. Despite the wariness of the managers in our company, we find that the price recommendation provided by the DST does serve as an effective reference price.

Our analysis shows that salespeople's decisions are well-explained by a two-stage decision model. Provided with a continuum of choices, we show that salespeople make an initial decision on whether or not to change a price (a binary decision) and then decide on the magnitude of change (a continuous response). The type of information most influential in the first stage varies from the information used in the second stage, indicating a hierarchy in information processing. We find that salespeople do differ in their reliance and receptiveness of pricing tool recommendations. In addition, we are able to identify cost-based windows of influence for the price recommendations. Our results have significant implications for the design of price optimization tools in settings where salespeople serve as the gatekeeper of price changes.

(Joint work with Wolfgang Jank, Itir Z. Karaesmen, and Shu Zhang)

"The Revenue Sharing in Airline Alliances"
Talk by Rene Caldentey
November 28, 2012
Airline alliances are a growing trend in the airline industry. From a revenue management perspective, one of the most significant features of these alliances are codeshare itineraries by which independent airlines can collaboratively market and operate flights. Different from traditional, monopolistic airline revenue management, alliance members control a decentralized network of resources through independent reservation and information systems.

In this research, we investigate contractual agreements between multiple airlines operating within a network alliance. We study contracts that specify how revenue should be split among the carriers. We propose a two-step hierarchical approach. We formulate a static problem in which airlines select partitioned allocations and show that a simple transfer price mechanism achieves first best. We study the dynamic problem and prove that the static transfer prices are asymptotically optimal.
(Joint work with Xing Hu, University of Oregon, and Gustavo Vulcano, NYU)

"The Dynamics of Power"
Talk by Sean Meyn, UIUC
September 18th, 2011
12:00-1:00 PM, room 561
We are moving towards a radical transformation of our energy systems. The success of the new paradigm created by the Smart Grid vision will require not only the creation and integration of new technologies into the grid, but also the redesign of its coupled market structures. Economic models able to capture the new physical reality are a first requirement for the design of a reliable, and "smart" electrical grid.

In a few years, smart meters and wind farms may be regarded as another "bridge to nowhere" unless we create the right architecture to make use of these resources.

To begin to address these issues, we survey elements of today's power grid, focusing on real time markets. While there are many success stories, the failures can be dramatic, as commonly seen recently in Australia, or in Texas last March, or in the midwest and New Zealand this past April. We investigate why these disasters occur, and conclude that they are a consequence of design: The static models used in competitive equilibrium analyses capture none of the key issues in real time markets. In particular, typical economic analyses ignore volatility on long and short time scales, and constraints due to the physical realities of the grid.

Several research questions are presented - their solution will require collaboration among researchers in economics, power and energy systems, and decision and control.

Based on the recent papers:
* A Control Theorist's Perspective on Dynamic Competitive Equilibria in Electricity Markets

* The Value of Volatile Resources in Electricity Markets

"Price Quoting Strategies of an Upstream Supplier"
Talk by Izak Duenyas, University of Michigan
September 28th, 2011
12:00-1:00 PM, room 561
This paper studies an upstream supplier who quotes prices for a key component to multiple sellers that compete for an end-buyer’s indivisible contract. At most one of the supplier’s quotes may result in downstream contracting and hence produce revenue for her. We characterize the supplier’s optimal price-quoting strategies and show that she will use one of two possible types of strategies, with her choice depending on the sellers’ profit potentials: secure, whereby she will always have business; or risky, whereby she may not have business. Addressing potential fairness concerns, we also study price-quoting strategies in which all sellers receive equal quotes. Finally, we show that the supplier’s optimal mechanism resembles auctioning a single quote among the sellers. This paper can assist suppliers at higher tiers of a supply chain in their pricing decisions, and provides general insights into multi-tier supply chains’ pricing dynamics.

"Online Market Places for Software Development Services – From Personal Experience to a Model of Buyer Choice in Multi-Attribute Auctions"
Talk by Christian Terwiesch, Wharton
October 5th, 2011
12:00-1:00 PM, room 561
This talk will begin with a description of two software development projects I have recently been involved in. Given by how much my software engineering knowledge is outdated by now, I decided to outsource most of the development work to professional coders. To identify capable coders, I used an online market place known as vworker.com (formerly known as rent-a-coder, similar to other such sites including odesk and elance) and received a fair number of bids. Bids submitted by coders to this site vary by price, coder knowledge, coder reputation, and geography, creating a multi-attribute decision problem. Based on several hundred thousand bids that I subsequently obtained from vworker over all auctions that had previously been hosted on their site, I will present a model of buyer choice. Particular emphasis will be given to the importance of coder reputation as well as to buyer preferences for geographic locations. This work is based on a collaboration with Antonio Moreno as well as Elena Krasnokutskaya.

"Fairness in Heavily-loaded Parallel Server Systems"
Talk by Mor Armony, NYU
October 12th, 2011
12:00-1:00 PM, room 561
Due to natural fluctuations in arrival rates and service times, many service systems experience heavily-loaded periods. Consequently, customers experience long delays and servers face high workloads. In such systems, it is desirable to maintain fairness with respect to a) the allocation of limited service resources among customers, and b) the division of workload among the servers. Patient flow data from an Israeli hospital indicate that both issues arise in reality. This talk will focus on fairness among customers. Specifically, we examine the behavior of parallel server queues during periods of overload. During such periods, the unstable queues may starve the limited resources. We explore the dynamics of the queue workloads under the well known MaxWeight scheduling policy during long periods of stress and discuss how to tune this policy in order to achieve a target fairness ratio across these workloads.
*Based on joint work with Nick Bambos and Carri Chan
*The empirical piece is based on joint work with Avi Mandelbaum, Yariv Marmor, Yulia Tseytlin and Galit Yom-Tov.

"Data-driven decision making with applications to healthcare systems"
Talk by Muhsen Bayati, Stanford GSB
October 19th, 2011
9:00-10:30 AM, room 561
Reconstructing a high-dimensional sparse vector from a small number of observations is a well-studied problem in many scientific, economic and engineering disciplines, and a number of tools have been designed to address this problem. It is currently experiencing a resurgence due to new applications, such as data-driven medicine and online advertising, and due to the need for accurate predictions under time and complexity constraints. This talk describes contributions in both directions.

In the first portion of this talk, I will describe the application of such tools for minimizing rehospitalizations--the admission of a patient to a hospital soon after discharge. Nearly one in every five patients is readmitted within 30 days of their discharge, and the estimated cost of such rehospitalizations to Medicare in 2004 was $17.4 billion. Hospitals aim to avoid rehospitalizations in a number of ways; for example, through patient education programs, follow-up home visits by pharmacists, and by supplying extensive discharge packages. It is important to properly allocate these costly and limited resources. Using electronic health records from a major hospital in the U.S., we have designed a predictive model which identifies patients with the highest risk of being rehospitalized, making it possible to significantly reduce rehospitalization costs.

In the second portion, I will focus on the rigorous analysis of a recent family of iterative algorithms for solving the above learning problems that are inspired by graphical models and ideas from statistical physics. These algorithms are exceptionally fast in yielding accurate predictions. Our analysis of these algorithms yields sharp formulas for their asymptotic performance. In particular, we derive rigorous formulas for the mean square error of the LASSO estimator.

"Making Better Fulfillment Allocation Decisions on the Fly"
Talk by Steve Graves, MIT
October 26th, 2011
9:15-10:45 AM, room 561
Online retailers manage large distribution networks, holding inventory in multiple locations to serve geographically-dispersed customers with varying service requirements. What is the best way to fulfill each customer's order when customers have different service-time requirements? We partner with a company that sells products online to examine this question, by first comparing a myopic strategy with a perfect hindsight optimization for a set of orders to characterize the potential improvement gap. We then report on the development of a heuristic that makes fulfillment decisions by minimizing the sum of the immediate cost plus an estimate of future expected opportunity cost of the decision. These estimates are derived from the dual values of a transportation problem assuming deterministic demand. In our experiments, we find that we can capture around 40% of the opportunity gap, leading to improvements on the order of 1%. (research with Jason Acimovic)

"Retail Assortment Optimization"
Talk by Marshall Fisher, Wharton
November 2nd, 2011
9:30-11:00 AM, room 561
A retailer’s assortment is the set of products they carry in each store in each point in time. Obviously the assortment a retailer chooses to carry has an enormous impact on their revenue, profitability and customer satisfaction. I describe joint work with Ramnath Vaidyanathan, a PhD student in Operations and Information Management at Wharton, now in the business school at McGill. We consider the problem of choosing, from a set of N potential SKUs in a retail category, K SKUs to be carried at each store of a retail chain so as to maximize sales or a defined profit function. Assortments can vary by store, subject to a maximum number of different assortments. We describe an approach in which we view a SKU as a set of attribute values, use sales history of the SKUs currently carried by the retailer to estimate the demand for attribute values and from this, the demand for any potential SKU, including those not currently carried by the retailer. We also introduce a model of substitution behavior, estimate the parameters of this model and consider the impact of substitution in choosing assortments. We use maximum likelihood estimation to fit the parameters of our model and describe several alternative heuristics for choosing SKUs. We describe application of this approach to optimize assortments for three real examples, snack foods, tires, and appearance chemicals. The tire and appearance chemicals recommendations were implemented and produced revenue increases of 5.8% and 3.5% respectively, increases which are significant relative to typical comparable store increases in these segments.

"Free Riding and Participation in Large Scale, Multi-Hospital Kidney Exchange"
Talk by Itai Ashlagi, MIT
November 9th, 2011
12:00-1:00 P.M., Room 561
As multi-hospital kidney exchange has grown, the set of players has grown from patients and surgeons to include hospitals. Hospitals can enroll only their hard-to-match patient-donor pairs, while conducting easily-arranged exchanges internally. This behavior has already been observed.

We show that the cost of making it individually rational for hospitals to participate fully is low in almost every large exchange pool (although the worst-case cost is very high), while the cost of failing to guarantee individual rationality could be high, in lost transplants. We identify a mechanism that achieves high efficiency while giving hospitals incentives to reveal all patient-donor pairs.

"Intensive Care Unit patient flow with readmissions: a state-dependent queueing network"
Talk by Carri Chen, Columbia GSB
November 30th, 2011
12:00-1:00 P.M., Room 561
Intensive Care Units (ICUs) provide care for the most critical patients in a hospital. Due to high staffing requirements and specialized equipment, they are very expensive to operate. As such, ICUs often operate at or above capacity, which leads to periods where patient demand exceeds availability. In such cases, current ICU patients may be discharged in order to accommodate new, more urgent patients. Such a discharge increases the likelihood of physiologic deterioration resulting in readmission to the ICU. We model such an ICU as a state-dependent Erlang-R queueing network where patient service times and readmission probabilities depend on the ‘overloaded’ or ‘underloaded’ state of the ICU. We refer to the reduction in Length-of-Stay due to incoming critical patients during overloaded periods as ‘speedup’. We consider how different definitions of ‘overload’ affect the steady-state behavior of the ICU and provide insight into capacity management of such systems. Using fluid models, we identify scenarios where speedup should never be used as well as scenarios where it can be helpful to temporarily alleviate congestion. We also look at patient flow data from 19 hospitals in a single hospital system and examine if and how speedup is utilized.

This talk is based on joint work with Galit Yom-Tov (Columbia University) and Gabriel Escobar (Kaiser Permanente)

back to top