The Yale Operations Seminar Series presents recent research papers in operations. The goal is to bring researchers from other universities to the Yale campus to stimulate exchange of ideas and deepen understanding of operations trends. These seminars are geared towards faculty and PhD students interested in operations research and management. Leading operations scholars, both internal and external will present their latest research. Doctoral students will meet with faculty prior to these seminars to review the papers and related literature. Participation in this seminar is required throughout the program.
Operations Seminar Series
Faculty Seminar Series Inaugurated on September 11, 2014
Faculty Coordinator: Associate Professor Vahideh Manshadi – 203.436.5026
Faculty Support Coordinator: Elizabeth Viloudaki – 203.436.5798
Tuesday Series 11:45 a.m.-12:45 p.m., 2020- via Zoom (formerly in Edward P. Evans Hall, 165 Whitney Avenue, New Haven, CT.)
An email notice with abstract and paper will be sent in advance of each talk in the series.
Current Semester – Spring 2021 (Virtual Seminars)
Vijay Kamble (Assistant Professor of Information and Decision Sciences, University of Illinois, Chicago), February 2
Individual Fairness in Hindsight (joint with Swati Gupta) The pervasive prevalence of algorithmic decision-making in societal domains necessitates that these algorithms satisfy reasonable notions of fairness. One compelling notion is that of individual fairness (IF), which advocates that similar individuals should be treated similarly (Dwork et al. 2012). In this paper, we extend the notion of IF to online contextual decisionmaking, in settings where there exists a common notion of conduciveness of decisions as perceived by the aected individuals. We introduce two denitions: (i) fairness-acrosstime (FT) and (ii) fairness-in-hindsight (FH). FT requires the treatment of individuals to be individually fair relative to the past as well as future, while FH only requires individual fairness of a decision at the time of the decision. We show that these two denitions can have drastically divergent implications in the setting where the principal needs to learn the utility model. Linear regret relative to optimal individually fair decisions is generally unavoidable under FT. On the other hand, we design a new algorithm: Cautious Fair Exploration (CaFE), which satises FH and achieves order-optimal sub-linear regret guarantees for a broad range of settings.
Forrest Crawford (Associate Professor of Biostatistics, Statistics and Data Science, Ecology and Evolutionary Biology, and Management, Yale University), February 16
Covid-19 Transmission Models in the Real World: Models, Data and Policy
Simple mathematical models of COVID-19 transmission gained prominence in the early days of the pandemic. These models provided researchers and policymakers with qualitative insight into the dynamics of transmission and quantitative predictions of disease incidence. More sophisticated models incorporated new information about the natural history of COVID-19 disease and the interaction of infected individuals with the healthcare system, to predict diagnosed cases, hospitalization, ventilator usage, and death. Models also provided intuition for discussions about outbreaks, vaccination, and the effects of non-pharmaceutical interventions like social distancing guidelines and stay-at-home orders. But as the pandemic progressed, complex real-world interventions took effect, people everywhere changed their behavior, and the usefulness of simple mathematical models of COVID-19 transmission diminished. This challenge forced researchers to think more broadly about empirical data sources that could help predictive models regain their utility for guiding public policy. In this presentation, I will describe my view of the successes and failures of population-level transmission models in the context of the COVID-19 pandemic. I will outline the evolution of a project to predict COVID-19 incidence in the state of Connecticut, from development of a transmission model to engagement with public health policymakers and initiation of a new data collection effort. In particular, I will argue that a new data source -- passive measurement of close interpersonal contact via mobile device location data -- is a promising way to overcome many of the shortcomings of traditional transmission models. I conclude with a summary of the impact this work has had on the COVID-19 response in Connecticut and beyond.
Chaithanya Bandi – March 2
David Alderson – March 23
Basak Kalkanci – April 6
Nikhil Garg – April 20
Antoine Desir – May 4
Sharad Goel (Assistant Professor of Management Science and Engineering, and by Courtesy, of Computer Science, Sociology and Law, Stanford University)
Designing Equitable Algorithms for Lending and Beyond
Machine learning methods are increasingly used to model risk in banking, criminal justice, healthcare, and beyond. These methods promise gains in accuracy, but also raise challenging statistical, legal, and other policy-relevant questions of fairness. In this talk, I’ll introduce and critique the dominant axiomatic approach to fairness in machine learning, arguing that common mathematical definitions of fairness can, perversely, lead to discriminatory outcomes in practice. I’ll then present an alternative perspective for designing equitable algorithms that foregrounds the inherent tension between competing concerns in real-world problems.
Hummy Song (Assistant Professor of Operations, Information and Decisions , University of Pennsylvania, Wharton)
Capacity Pooling in Hospitals: The Hidden Consequences of Off-Service Placement
In order to better manage the supply-demand mismatches between bed types and patient types, many hospitals engage in capacity pooling by assigning patients from a service with limited or no available beds to an available bed in a unit designated for a difference service. This “off-service placement” is a common occurrence in many hospitals, yet its effects are not well understood, perhaps because of the endogenous selection of off-service patients. In the first paper, we quantify the effect of off-service placement on those patients who have been placed off service; we find that they experience a 25% increase in length of stay and also an increase in likelihood of readmission. In the second paper, we examine the extent to which patients who are placed on service are impacted by the service’s off-service activities. Our findings suggest that this spillover effect is substantial, and that both the level and the volatility of off-service placement impacts those patients who are placed on service. Quantifying these various effects of off-service placement can inform future models of patient flow and enable hospitals to make better informed decisions around capacity allocation and off-service placement.
Elisa Long (Assistant Professor of Decisoins, Operations, and Technology Management, UCLA, Anderson School of Management)
Nursing Home Staff Newtorks and Covid-19
Nursing homes and other long-term care facilities account for a disproportionate share of COVID-19 cases and fatalities worldwide. Outbreaks in U.S. nursing homes have persisted despite nationwide visitor restrictions beginning in mid-March. An early report issued by the Centers for Disease Control and Prevention identiﬁed staﬀ members working in multiple nursing homes as a likely source of spread from the Life Care Center in Kirkland, Washington to other skilled nursing facilities. The full extent of staﬀ connections between nursing homes—and the role these connections serve in spreading a highly contagious respiratory infection—is currently unknown given the lack of centralized data on cross-facility employment. We perform the ﬁrst large-scale analysis of nursing home connections via shared staﬀ and contractors using device-level geolocation data from 50 million smartphones, and ﬁnd that 5.1 percent of smartphone users who visit a nursing home for at least one hour also visit another facility during our 11-week study period—even after visitor restrictions were imposed. We construct network measures of connectedness and estimate that nursing homes, on average, share connections with 7 other facilities. Controlling for demographic and other factors, a home’s staﬀ-network connections and its centrality within the greater network strongly predict COVID-19 cases. Traditional federal regulatory metrics of nursing home quality are unimportant in predicting outbreaks, consistent with recent research. Multivariate regressions comparing demographically and geographically similar nursing homes suggest that 49 percent of COVID cases among nursing home residents are attributable to staﬀ movement between facilities.
Yao Cui (Assistant Professor of Operations, Technology and Management, Cornell University)
Tax Induced Inequalities in the Sharing Economy
The growth of sharing economy marketplaces like Airbnb has generated discussions on their socioeconomic impact and lack of regulation. As a result, most major cities in the United States have started to charge an “occupancy tax” (which is common for hotels) to Airbnb bookings. In this study, we investigate the eﬀects of the occupancy tax policy on Airbnb listings, using a combination of a generalized causal forest methodology and a diﬀerence-in-diﬀerences framework. Further, given the considerable variety among listings on Airbnb, we also estimate the heterogeneous treatment eﬀects of the tax. One key ﬁnding is that the tax reduces listing revenues substantially, but Airbnb hosts are reluctant to respond by reducing the listing prices. A second important result is that the tax policy is more favorable to commercial hosts with multiple properties or entire-space (“target”) listings, versus residential hosts with single shared-space (“non-target”) listings. We further show evidence that this heterogeneous treatment eﬀect is caused by customers’ discriminatory tax aversion against residential hosts. Stemming from these empirical results we prescribe how hosts should optimally set prices in response to the occupancy tax and also identify the discriminatory tax rates that would equalize the tax’s eﬀect across target and non-target listings.
Simone Marinesi (Assistant Professor of Operations, Information and Decisions, University of Pennsylvania, Wharton)
Rethinking Crowdfunding Platform Design: Mechanisms to Deter Misconduct and Improve Efﬁciency
Lacking credible rule-enforcement mechanisms to punish misconduct, existing reward-based crowdfunding platforms can leave backers exposed to two risks: entre-preneurs may run away with backers’ money (funds misappropriation), and product speciﬁcations may be misrepresented (performance opacity). We show that each of these risks can materially impact crowdfunding efﬁciency, and, when jointly present, they in-teract with each other in ways that can dampen or, more worryingly, amplify their in-dividual adverse effects. To mitigate these risks, we propose two mechanisms based on deferred payments. The ﬁrst involves stopping the campaign once the funding goal is reached and servicing any unmet demand in the aftermarket. The second involves escrowing any funds raised in excess of the goal, as insurance for backers. We show that early stopping dominates escrow and boosts platform revenues. Pairing these deferred payment designs with (costly) performance veriﬁcation contingencies can bring additional gains, but doing so can ﬂip their relative performance, with escrow coming out on top. Overall, by accounting for different timing (pre- versus post-campaign) and enforcement rules (mandatory versus optional) of the veriﬁcation contingencies, we analyze a total of 10 different designs and show that two of them dominate: the early stopping design and the escrow design with mandatory ex-post veriﬁcation. We conclude by providing recom-mendations for which design works best under different conditions and exploring the potential of crowdsourced performance checks.
Yaron Shaposhnik (Assistant Professor, Simon Business School, University of Rochester)
Globally-Consistent Rule-based Summary Explanation for Machine Learning Models: Application to Credit-Risk Evaluation
We develop a method for interpreting specific predictions made by (global) predictive models by constructing (local) models tailored to each specific observation (these are also called ``explanations" in the literature). Unlike existing work that ``explains'' specific observations by approximating global models in the vicinity of these observations, we fit models that are globally-consistent with predictions made by the global model on past data. We focus on rule-based models (also known as association rules or conjunctions of predicates), which are interpretable and widely used in practice. We design multiple algorithms to extract such rules from discrete and continuous datasets, and study their theoretical properties. Finally, we apply these algorithms to multiple credit-risk models trained on real-world data from FICO and demonstrate that our approach effectively produces sparse summary-explanations of these models in a short period of time. Our approach is model-agnostic (that is, can be used to interpret any predictive model), and solves a minimum set cover problem to construct its summaries.
Shouqiang Wang (Assistant Professor, Operations Management, University of Texas at Dallas, Naveen School of Management)
Engineering Social Learning: Information Design of Time-Locked Sales Campaigns for Online Platforms
Many online platforms (e.g., Amazon and Groupon) oﬀer time-locked sales campaigns as an innovative sell-ing mechanism, whereby third-party vendors sell their products at a ﬁxed price for a pre-speciﬁed length of time. To alleviate customers’ uncertainty and to inﬂuence their inference about a product’s value, platforms often display to upcoming customers some information about previous customers’ purchase decisions, which are the platform’s proprietary observation. Using a dynamic Bayesian persuasion framework, we formulate and study how a platform should optimally design its dynamic information provision strategy to maximize its expected revenue. We establish an equivalent reformulation of the platform’s information design prob-lem by signiﬁcantly reducing the dimensionality of the platform’s message space and proprietary history. Speciﬁcally, we show that it suﬃces for the platform to use only three messages in disclosing information: a neutral recommendation that induces a customer to make her purchase decision according to her private assessment about the product; and a positive (resp., negative) recommendation that induces a customer to make the purchase (resp., not to make the purchase) by ignoring her private assessment. We also show that the platform’s proprietary history can be represented by the net purchase position, a single-dimensional summary statistic that computes the diﬀerence between the cumulative purchases and non-purchases made by customers who receive the neutral recommendation from the platform. Subsequently, the platform’s prob-lem can be formulated and solved eﬃciently as a linear program. Further, we propose and optimize over a class of heuristic policies. The best heuristic policy, which we characterize analytically, is easy-to-implement, simple-to-prescribe, and near-optimal policy. Speciﬁcally, this heuristic policy provides only neutral recommendations to customers arriving up to a cut-oﬀ customer and provides only positive or negative recommendations to customers arriving afterwards. The recommendation is positive if and only if the net purchase position achieved right after the cut-oﬀ customer exceeds a threshold. Finally, we demonstrate that the best heuristic policy improves the platform’s revenue over naıve policies commonly used in practice, such as the no-disclosure and full-disclosure policies, and captures at least 90% of the optimal revenue.
David Shmoys (Laibe/Acheson Profesor of Business Management and Leadership, Department of Computer Science, Cornell)
Models and Algorithms in Support of Modern Urban Mobility
-joint with Daniel Freund, Shane Henderson and Eoion O’Mahony
The sharing economy is transforming how people move within cities. Ridesharing, bike-sharing (both station-based and dockless, with both traditional and electric pedal-assist bikes), and scooters have augmented the options present for decades in terms of public transit and private vehicle ownership. We will discuss the issues raised by this transformation, while focusing on some of the operational challenges present in station-based bike-sharing systems. We have worked with Citibike (the operator of bike-sharing in NYC) and its then parent company Motivate, developing optimization models and algorithms to change how they manage their systems. For example, continuous-time Markov chain models, combined with simple mathematical programming tools, can be used to answer the question – what is the optimal deployment of the bike fleet across the system at the start of the day? Furthermore, we consider the more strategic question of how to (re-) allocate dock-capacity in such systems, which gives rise to new (solvable) integer programming problems. We have also guided the development of Bike Angels, a program to incentivize users to crowdsource “rebalancing rides”; we will describe its underlying analytics, where the pricing mechanism is once again grounded in the same underlying algorithmic tools. Finally, we will discuss the range of further opportunities raised by the evolving landscape in urban mobility.
Jeho Lee (Professor of International Business, Graduate School of Business, Seoul National University)
In Search of Robust Design Principles: Anomalies in the Hub-and-Spoke Networks of Major US Carriers
In the hub-and-spoke networks of major U.S. carriers, a triad represents three airports that are fully connected through direct routes by operation of nonstop flights, which is costly to maintain. However, triads are far more pervasive than might be expected given the expense. Why do the carriers maintain so many triads? Our analysis reveals that triads cushion airline services against unexpected flight cancelations and facilitate travel via reasonable alternative routes. The operational resilience provided by triads stems from the fact that 99% of triads are concentrated in subnetworks surrounding hub airports. This overconcentration of triads around a few hubs deviates substantially from the topological properties predicted by the well-known models of hubs. Our work sheds new light on the intricacies associated with the costs and benefits of triads in developing routes.
Kimon Drakopoulos (Assistant Professor of Data Sciences and Operations, University of Southern California)
Why Perfect Tests May Not be Worth Waiting For: Information as a Commodity
-joint with R.S. Randhawa
Information products provide agents with additional information that is used to update their actions. In many situations, access to such products can be quite limited. For instance, in epidemics, there tends to be a limited supply of medical testing kits, or tests. These tests are information products because their output of a positive or a negative answer informs individuals and authorities on the underlying state and the appropriate course of action. In this paper, using an analytical model, we show how the accuracy of a test in detecting the underlying state serves as a rationing device to ensure that the limited supply of information products is appropriately allocated to the high demand by heterogeneous agents. On the technical side, we ﬁnd that in many settings, providing perfect information (or a perfect test) is sub-optimal, and a moderately good test is preferable. On the policy side, we use a numerical study of an evolving epidemic to conﬁrm our theoretically derived insight that in the early stages of an epidemic investing on higher testing quality is not beneﬁcial if testing availability is low.
Hamsa Sridhar Bastani (Assistant Professor of Operations, Information and Decisions, Wharton School of Business)
Predicting with Proxies: Transfer Learning in High Dimension
Despite the big data revolution, "small data" problems are ubiquitous in healthcare, marketing, and pricing. Transfer learning is a promising approach to improve prediction accuracy by incorporating data from closely related proxy outcomes. For example, e-commerce platforms use abundant customer click data (proxy) to make product recommendations rather than the relatively sparse customer purchase data (true outcome of interest); alternatively, hospitals often rely on medical risk scores trained on a different patient population (proxy) rather than their own patient population (true cohort of interest) to assign interventions. Yet, not accounting for the bias in the proxy can lead to sub-optimal decisions. Using real datasets, we find that this bias can often be captured by a sparse function of the features. Thus, we propose a novel two-step estimator that uses techniques from high-dimensional statistics to efficiently combine a large amount of proxy data and a small amount of true data. We prove upper bounds on the error of our proposed estimator and lower bounds on several heuristics used by data scientists; in particular, our proposed estimator can achieve the same accuracy with exponentially less true data. We demonstrate the effectiveness of our approach on e-commerce and healthcare datasets; in both cases, we achieve significantly better predictive accuracy as well as managerial insights into the nature of the bias in the proxy data.
Oguzhan Alagoz (Visiting Professor of Operations at Kellogg School of Management and Proctor and Gamble Bascom Professor of Industrial Systems Engineering, University of Wisconsin-Madison)
A Mathematical Modeling Framework to Personalize Mammography Screening Decisions
Breast cancer is the most common non-skin cancer and the second leading cause of cancer-death in US women. Although mammography is the most effective modality for breast cancer diagnosis, it has several potential risks, including high false positive rates, which are not very rare. Therefore, the balance of benefits and risks, which depend on personal characteristics, is critical in designing a mammography screening schedule. In contrast to prior research and existing guidelines which consider population-based screening recommendations, we propose a personalized mammography screening policy based on the prior screening history and personal risk characteristics of women. We formulate a finite-horizon partially observable Markov decision process (POMDP) model for this problem. Our POMDP model incorporates two methods of detection (self or screen), age-specific unobservable disease progression, and age-specific mammography test characteristics. We use a validated micro-simulation model based on real data in estimating the parameters and solve this POMDP model optimally for individual patients. Our results show that our proposed personalized screening schedules outperform the existing guidelines with respect to the total expected quality-adjusted life years, while significantly decreasing the number of mammograms. We further find that the mammography screening threshold risk increases with age. We derive several structural properties of the model, including the sufficiency conditions that ensure the existence of a control-limit policy. Finally, we briefly describe our findings using two extensions to the basic POMDP model. The first extension personalizes breast cancer screening while accounting for the nonadherence of women to screening recommendations and the second extension considers allocating limited breast cancer screening resources, a common problem in developing countries and resource-limited settings.
Florin Ciocan (Assistant Professor of Technology and Operations Management, INSEAD)
Interpretable Optimal Stopping
Optimal stopping is the problem of deciding when to stop a stochastic system to obtain the greatest reward, arising in numerous application areas such as finance, healthcare and marketing. State-of-the-art methods for high-dimensional optimal stopping involve approximating the value function or the continuation value, and then using that approximation within a greedy policy. Although such policies can perform very well, they are generally not guaranteed to be interpretable; that is, a decision maker may not be able to easily see the link between the current system state and the policy's action. In this paper, we propose a new approach to optimal stopping, wherein the policy is represented as a binary tree, in the spirit of naturally interpretable tree models commonly used in machine learning. We formulate the problem of learning such policies from observed trajectories of the stochastic system as a sample average approximation (SAA) problem. We prove that the SAA problem converges under mild conditions as the sample size increases, but that computationally even immediate simplifications of the SAA problem are theoretically intractable. We thus propose a tractable heuristic for approximately solving the SAA problem, by greedily constructing the tree from the top down. We demonstrate the value of our approach by applying it to the canonical problem of option pricing, using both synthetic instances and instances calibrated with real S&P-500 data. Our method obtains policies that (1) outperform state-of-the-art non- interpretable methods, based on simulation-regression and martingale duality, and (2) possess a remarkably simple and intuitive structure.
John Birge (Jerry W. and Carol Lee Levin Distinguished Service Professor of Operations Management, University of Chicago, Booth)
Dynamic Learning in Strategic Pricing Games
In monopoly pricing situations, firms should optimally vary prices to learn demand. The variation must be sufficiently high to ensure complete learning. In competitive situations, however, varying prices provides information to competitors and may reduce the value of learning. Such situations may arise in the pricing of new products such as pharmaceuticals and digital goods. This paper shows that firms in competition can learn efficiently in certain equilibrium actions which involve adding noise to myopic estimation and best-response strategies. The paper then discusses how this may not be the case when actions reveal information quickly to competitors. The paper provides a setting where this effect can be strong enough to stop learning so that firms optimally reduce any variation in prices and choose not to learn demand. The result can be that the selling firms achieve a collaborative outcome instead of a competitive equilibrium. The result has implications for policies that restrict price changes or require disclosures.
Foad Iravani (Assistant Professor of Operations Management, University of Washington, Foster)
A Reduce-To-Threshold Approach to Direct Load Control Contracts with Monthly Constraints
Balancing demand and supply of electricity is one of the most important tasks that utility firms perform on a daily basis to maintain grid stability and reduce system cost. Demand response programs are among strategies that utilities use to reduce electricity consumption during peak hours. Direct Load Control Contracts (DLCC) are a class of incentive-based demand response programs that allow utilities to directly “call” consumers through remote control devices and reduce their energy usage. Given the rapid expansion of such contracts in practice, in this paper we formulate an optimization model for executing DLCCs that minimizes total system cost subject to monthly and annual constraints on the number of times and hours customers can be called. Given the complex structure of the problem, we develop a hierarchical approximation scheme that solves the problem effectively. Our experiments with real data from the California system operator show that our approximation scheme performs remarkably well and can reduce the total cost by 3.81% on average and as much as 6.31%.
Karl Sigman (Professor of Industrial Engineering and Operations Research, Columbia University)
Exact/Perfect Simulation with Applications in Operations Research
An introduction and overview of a simulation method known as "Exact" or "Perfect" Simulation is first given, with some simple motivating examples, some going back to John van Neumann. Then, in the context primarily of queueing theory/Operations Research, some more detailed examples and results are given such as how to exactly simulate from the limiting (stationary) distribution of some well known queueing models such as muliti-server queues with renewal arrivals and independent identically distributed service times. More recent work is discussed, and then finally some ongoing research on applications to complex networks of queues is discussed with its challenges, and open problems.
Karen Smilowitz (Industrial Engineering and Management Sciences, Northwestern University)
Leveraging network structure in covering path problems: An application to school bus routing
In the face of critical budget cuts, many public school districts are looking to reduce transportation expenditures and keep cuts away from the classroom. Our research team at Northwestern University has been exploring such challenges in partnership with Evanston / Skokie District 65, a preK-8 public school district north of Chicago. In this talk, I will provide a broad overview of the partnership and present our initial work on the School Bus Routing Problem (SBRP). The SBRP has been studied by the operations research community for over fifty years, identifying creative routing and scheduling approaches for school districts. The SBRP itself is a composite of five decision sub-problems: data preparation, bus stop selection, bus route generation, school bell time adjustment, and route scheduling. As a first step towards a comprehensive solution approach, we introduce the covering path problem on a grid (CPPG) which finds a cost-minimizing path connecting a subset of points in a grid such that each point is within a predetermined distance of a point from the chosen subset. We leverage the geometric properties of the grid graph which captures the road network structure in many transportation problems, including our motivating setting of school bus routing. With this network structure, we are able to develop efficient methods to find feasible, high quality solution paths for the CPPG. Our results for the stylized grid setting establish important building blocks for more general settings.
Roman Kapuscinski (John Psarouthakis Research Professor of Manufacturing Management, University of Michigan)
Strategic Behavior of Suppliers in the Face of Production Disruptions
To mitigate supply disruption risks, some manufacturers adopt a flexible sourcing strategy, where they have an option of sourcing from multiple suppliers and assigning different roles (regular and backup) to the suppliers. The paper evaluates the costs and benefits associated with such a flexible sourcing plan when suppliers are strategic price setters. It finds that suppliers’ strategic behavior dramatically changes the outcome. Rather than promoting the suppliers’ competition, the flexible-sourcing plan may incentivize the suppliers to compete less aggressively.
Minimum Advertised Price Policy: Economic Analysis and Implications
During last twenty years, many brick-and-mortar retailers face competition from online retailers and local discounters. Customers are able to experience products in a brick-and-mortar store but purchase online for lower prices. As a result, brick-and-mortar retailers' sales decrease and they stop promoting or carrying such products. For manufacturers, however, brick-and-mortar retailers play a crucial role by showcasing and advertising products to customers. Resale Price Maintenance (RPM) and Minimum Advertised Price (MAP) are two commonly-used policies intended to protect retailers' margin. In this paper, we build a stylized model to study RPM and MAP under various market situations. In particular, we explore which policy is more beneficial for the manufacturer, retailers and consumers.
Martin Lariviere (John L. and Helen Kellogg Professor of Operations, Kellogg School of Business, Northwestern University)
Priority Queues and Consumer Surplus
We examine whether priority queues benefit or hurt customers in a setting in which customers are privately informed of their per-unit-time waiting cost. Implementing a priority queue thus means posting a menu of expected waits and out-of-pocket prices that are incentive compatible. Whether priorities increase or decrease consumer surplus relative to first-in, first-out service depends on the model of customer utility and on the distribution of customer waiting costs. If all customers have the same value of the service independent of their waiting costs, priorities essentially always lower consumer surplus. If a customer’s value of the service is an increasing function of her waiting cost, priorities lower surplus if the distribution of waiting costs has a decreasing mean residual life. If the mean residual life is increasing, then priorities make consumers better off. We show that the results across utility models are linked by an elasticity measure. If an appropriate measure of waiting cost is elastic, consumer surplus falls with priorities. We also explore how priorities impact individual customers and show that they potentially make all customers worse off. It is possible that low priority customers may pay a higher out-of-pocket price than they would under first-in, first-out service.
Jose Guajardo (Haas School of Business, University of California, Berkeley)
How Do Usage and Payment Behavior Interact in Rent-to-Own Business Models? Evidence from Developing Economies
The diffusion of technological innovations in developing economies has been facilitated by the use of rent-to-own business models, which give flexibility to customers by allowing them to make incremental payments over time. Understanding the implications of this flexibility is a fundamental problem for an increasing number of firms operating in these markets. In this paper, we empirically analyze how consumer usage and payment behaviors interact in an application of rent-to-own to the distribution of solar lamps in developing countries. By exploiting the longitudinal variation in the data—and hence accounting for intrinsic differences between customers—the analysis led to three main insights. First, higher usage rates lowered the probability of late payments by customers. Our characterization of this engagement effect enhances existing knowledge of the drivers of payment behavior in these environments. Second, customers often “bundled” payments, making advance payments for future product access. We showed that bundling the initial payment led to lower usage rates (bundling effect), suggesting that firms may not benefit from advance payments upfront, and that they should closely track usage patterns from these customers. Finally, we showed that first-period usage information can improve the accuracy of predictive models of default and that observing usage rates in subsequent periods does not lead to further improvements. Overall, the analysis highlights the importance for firms of jointly tracking and analyzing payment and usage behavior by customers, particularly in initial stages of the adoption process.
Mohamed Mostagir (Assistant Professor of Technology and Operations, Ross School of Business, University of Michigan)
Dynamic Contest Design: Theory, Experiments, and Applications
Contests are a common mechanism for extracting effort from participants. Their use is widespread in a variety of settings like workplace promotions, crowdsourcing innovation, and healthcare quality. One of the pivotal aspects of contest design is the contest's information structure: what information should the contest designer provide to participants and when should this information be revealed? The answers to these questions have important consequences on the behavior of players and the outcome of the contest, as well as implications for institutional and policy design. We derive the contest's optimal information disclosure policy within a large class of policies and design a novel experiment to evaluate how these policies perform in the lab.
Junior Operations Conference
February 8 and 9, 2018
- Maria Ibanez (Harvard)- “Discretionary Task Ordering: Queue Management in Radiological Services”
- Ali Makhdoumi (MIT)- “Fast and Slow Learning from Reviews”
- Irene Lo (Columbia)- “Dynamic Matching in School Choice: Efficient Seat Reassignment after Late Cancellations”
- Daniel Freund (Cornell)- “Models and Algorithms for Transportation in the Sharing Economy”
- Chloe Glaeser (Wharton)- “Optimal Retail Location: Empirical Methodology and Application to Practice”
- Joann De Zheger (Stanford)- “Pay It Forward: Sustainability in Smallholder Commodity Supply Chains”
Marcelo Olivares (Industrial Engineering, Universidad de Chile)
Managing Worker Utilization in Service Platforms: An Empirical Study of an Outbound Call-Center
-joint with Andres Musalem and Daniel Yung
In many service industries, providing prompt response to customers can be an important competitive advantage, especially when customers are time-sensitive. When demand for the service is variable and the staffing requirements cannot be adjusted quickly, capacity decisions require making a trade-off between the responsiveness to customers versus controlling operating costs through worker utilization. To break this trade-off, the service system can operate as a platform with access to a large pool of employees with flexible working hours that are compensated through piece-rates. Examples of these service platforms can be found in transportation, food delivery and customer contact centers, among many others. While this business model can operate at low levels of utilization without increasing operating costs, a different trade-off emerges: in settings where employee training and experience is important, the service platform must control employee turnover, which may increase when employees are working at low levels of utilization. Hence, to make staffing decisions and managing workload, it is necessary to understand both customer behavior (measuring their sensitivity to service times) and employee retention. We analyze this trade-off in the context of an outbound call-center that operates with a pool of flexible agents working remotely, selling auto insurance. We develop an econometric approach to model customer behavior in the context of an out-bound call center, that captures special features of out-bound calls, time-sensitivity and the effect of employee experience. A survival model is used to measure how agent retention is affected by the assigned workload. These empirical models of customers and agents are combined to illustrate how to balance time-sensitivity and employee experience, showing that both effects are relevant in practice to plan workload and staffing in a service platform.
Peter Kolesar (Professor Emeritus, Columbia University, Member, Columbia Water Center)
Breaking the Deadlock: Improving Water-Release Policies on the Delaware River through Operations Research
The Delaware River provides half of New York City's drinking water and impacts water supply for Philadelphia and central New Jersey. The upper river is a habitat for wild trout and American shad and has suffered three 100-year floods in recent years. The water releases from three New York City dams on the Delaware River's headwaters impact the reliability of the city’s water supply, the potential for downriver floods, and the quality of the aquatic habitat. We describe the efforts of an environmental coalition to revise the Delaware’s release policies to benefit river habitat and fisheries without increasing the New York City's drought risk. Changes in the release policies were constrained by the dictates of a 1954 US Supreme Court decree and the 1961 multi-state/federal Delaware River Basin Compact which mandated unanimity among the states of New York, New Jersey, Pennsylvania and Delaware -- and New York City. These five parties to the 1954 Supreme Court decree have sometimes conflicting water objectives. We describe the analyses, activism and the politics that led to the Delaware River Basin Commission to implement the Flexible Flow Management Program (FFMP), an operations research-based water release framework in October 2007, and substantially improve it in 2017. In addition to meeting its habitat improvement goals and drought- risk constraint, the FFMP algorithm modestly increases flood mitigation and is substantially simpler to administer. We will discuss the recent dispute between New York City and Jew Jersey over future water rights, and our OR based proposals for further improvements in the FFMP.
Yash Kanoria (Associate Professor, Decision, Risk and Operations Division, Columbia Business School, Columbia University)
State Dependent Control of Ride Hailing Systems
-joint with Sid Banerjee and Pengyu Qian
Inspired by ride-hailing and bike-sharing systems, we study the design of state-dependent controls for a closed queueing network model. We focus on the assignment policy, where the platform can choose which nearby vehicle to assign to an incoming customer; if no units are available nearby, the request is dropped. The vehicle becomes available at the destination after dropping the customer. We study how to minimize the proportion of dropped requests in steady state. We propose a family of simple state-dependent policies called Scaled MaxWeight (SMW) policies that dynamically manage the geographical distribution of supply. We prove that under the complete resource pooling (CRP) condition (analogous to the condition in Hall's marriage theorem), each SMW policy leads to exponential decay of demand-dropping probability as the number of supply units scales to infinity. Further, there is an SMW policy that achieves the *optimal* exponent among all assignment policies, and we analytically specify this policy in terms of the customer arrival rates for all source-destination pairs. The optimal SMW policy maintains high supply levels near structurally under-supplied locations. We also propose data-driven approaches for designing SMW policies and demonstrate excellent performance in simulations based on the NYC taxi dataset.
Philipp Afèche (Associate Professor of Operations Management, Rotman School of Management, University of Toronto)
Ride-Hailing Networks with Strategic Drivers: The Impact of Platform Control Capabilities on Performance
-joint with Costis Maglaras and Zhe Liu. (The talk is based on a paper that is currently being revised.)
Motivated by ride-hailing platforms such as Uber and Lyft, we study the problem of matching demand (riders) with self-interested capacity (drivers) over a spatial network. We focus on the performance impact of two operational platform control capabilities, demand-side admission control and supply-side repositioning control, considering the interplay with two practically important challenges: Significant spatial demand imbalances prevail for extended periods of time; and drivers are self-interested and strategically decide whether to join the network, and if so, when and where to reposition when not serving riders. We obtain the following results in the context of a game-theoretic fluid model of a two-location, four-route network. 1. We characterize and compare the steady-state system equilibria under three control regimes, ranging from minimal platform control to centralized admission and repositioning control. 2. We provide a new result on the interplay of platform admission control and drivers’ repositioning decisions: It may be optimal to strategically reject demand at the low-demand location even if drivers are in excess supply, to induce repositioning to the high-demand location. 3. We derive tight upper bounds on the platform’s and the drivers’ gains due to platform controls; these gains are more significant under moderate capacity and significant cross-location demand imbalance.
Andre Calmon (Assistant Professor of Technology and Operations Management and the Patrick and Valentine Firmenich Fellow for Business and Society at INSEAD)
Operations Strategy at the Base of the Pyramid: Consumer Education and Reverse Logistics in a Durable Goods Supply Chain
We model and analyze the value of different operations strategies for a social enterprise that distributes durable life-improving technologies to low-income consumers in the Base of the Pyramid (BOP). This distributor considers two strategies: (i) improved consumer education and (ii) improved reverse logistics. We prove that (ii) is always beneficial for the distributor, while (i) may be detrimental. Moreover, we show that this effect is weakened (strengthened) if the consumers' financial distress is high (low) and the distributor highly values product adoptions relative to profits. We use this framework to contrast the value of strategies (i) and (ii) in the BOP for non-profit organizations that primarily value adoption of life-improving products, social enterprises that value both profit and adoption, and for-profit businesses that primarily value profit. We compare these strategies with those in high-income markets where consumers do not suffer extreme financial distress.
Shane G. Henderson (Professor and Director, School of Operations Research and Information Engineering)
Citi Bike Planning
-joint with Daniel Freund, Nanjing Jian, Eoin O’Mahony and David Shmoys
Cornell’s School of Operations Research and Information Engineering has been working with the bike-sharing company Citi Bike since Citi Bike began operations in New York City in 2013. We provide data analysis and advice about strategy and operations, not just to Citi Bike, but also to its parent company Motivate that operates several bike-sharing programs around the USA. I will describe some of our modeling work with Citi Bike, focusing on a suite of models (not just simulation models) that informs the decision about where to position both racks and bikes around the approximately 750 stations in NYC.
Rene Caldentey (Professor of Operations Management, The University of Chicago , Booth)
On the Optimal Design of a Bipartite Matching System
-joint with Philipp Afeche (U. Toronto) and Varun Gupta (U Chicago)
In this talk, we explore the optimal design of matching topologies for a multi-class multi-server queueing system in which each customer class has a specific preference over server types. We investigate the performance of the system from the perspective of a central planner who must decide the set of feasible customer-server pairs that can be matched together under fairness constraints for both customers and servers.
Johannes Ledolter (Professor of Management Sciences and Statistics & Actuarial Scince, Robert Thomas Holmes Professor, The University of Iowa, Tippie College of Business)
Data Science Projects: Treatment of Visual Loss and Effects of Traumatic Brain Injury
In this talk I review and evaluate useful methods in data science. Data science helps us understand the information in both big and small data sets. In business applications, data science helps us gain a competitive advantage; targeted marketing, made possible by the analysis of millions of transaction data, is a prime example. Data science also helps physicians develop efficient personalized medical treatments and helps researchers understand the links between concussions and blast exposure and injuries to the brain and the visual system. I have been involved in studies from ophthalmology where data science methods have proved useful. These studies focus on the thickness of the retinal nerve fiber layer. The analysis of the thickness in healthy eyes and diseased eyes helps with diagnosis and provides treatment guidance for glaucoma, multiple sclerosis and retinal diseases such as age-related macular degeneration and diabetic eye disease. Thickness measurements also reflect the effects of concussions and traumatic brain injury. Athletes and members of the military are prime candidates for traumatic brain injury. Many deceased NFL players have chronic traumatic encephalopathy (CTE), a degenerative brain disorder associated with repetitive head trauma. Our research indicates that the effects of concussions and blast exposure can be detected earlier in the optic nerve (the gateway to the brain) and the thickness of the retinal nerve fiber layers. I discuss two studies and describe the statistical methods used. The first study involves nerve layer thickness comparisons of glaucoma and normal subjects. The second study on Iowa football players explores whether differences in the thickness of the retinal nerve fiber layers are related to player positions.
Paat Rusmevichientong (Professor, Data Sciences and Operations, University of Southern California, Marshall School of Business)
A New Approach in Approximate Dynamic Programming for Revenue Management of Reusable Products)
-joint with Huseyin Topaloglu and Mika Sumida (Cornell Tech)
We present a new approach in approximate dynamic programming for revenue management of reusable products. The problem is motivated by emerging industries that rent out computing capacity and fashion items, where customers request products on-demand, use the products for a random duration of time, and afterward return the products back to the firm. The goal is to find a policy that determines what products to offer to each customer to maximize the total expected revenue over a finite selling horizon. For this problem, the firms must simultaneously consider the inventories of available products, along with the products that are currently in use by other customers. So, the resulting dynamic programming formulation is intractable because of the high-dimensional state variable.Using a novel approach for constructing an affine approximation to the value functions, we present a policy that is guaranteed to obtain at least 50% of the optimal expected revenue. Our construction is based on a simple and efficient backward recursion. We provide computational experiments based on the parking transaction data in Seattle. Our numerical experiments demonstrate that the practical performance of our policy is substantially better than its worst-case performance guarantee
Hamed Mamani (Associate Professor of Operations Management, Premera Endowed Professor, University of Washington, Foster School of Business)
Payment Models in Healthcare
Healthcare spending accounts for 17.5% of the GDP in the United States and has been rising year after year, which makes this a significant concern for society. There have been many proposals specifically related to the reimbursement models from insurers to providers to curb the healthcare cost. However, many of these proposals are not studied or thoroughly evaluated before they are implemented. To this extent, we consider two healthcare payments: bundled payment (BP) and reference pricing (RP), where we investigate the current policies in healthcare payments, propose new payments that can better align the incentives of the stakeholders to the system optimum outcomes, compare and contrast different payment systems, and support the derivations with empirical evidence. The current reimbursement system for hospitals is mainly fee-for-service (FFS), under which the hospital is reimbursed for every procedure, test, etc. This payment model motivates the hospitals to treat patients more than is needed and to perform unnecessary tests that contribute to excessive healthcare expenditures. There have been many proposals for alternative payment models especially from the Center for Medicare and Medicaid services (CMS). One way of controlling healthcare expenditures is by modifying the reimbursement schemes to share some of the insurers’ risk with providers. In the first part of this talk we consider bundled payment, which is a prime example of such policies. The second part of the talk is related to impacting healthcare spending through cost-sharing mechanisms with patients, which affects the providers’ market shares and, subsequently their pricing strategies. We study reference pricing in this area, and questions that we are trying to address are: what is the impact of RP and if it can achieve its premises or are there unintended consequences?
Pinar Keskinocak (Willam W. George Chair and ADVANCE Professor and Interim Associate Dean for Faculty Development & Scholarship, College of Enginering and Director of the Center for Health and Humanitarian Systems, Georgia Tech Steward School of Industrial and System Engineering)
Quantitative Models Embedded in Decision-Support Tools for Healthcare Applications
We will discuss a few examples from healthcare, where quantitative models embedded in decision-support tools can improve the quality of decision-making (for patients, physicians, or caregivers) and patient outcomes. Examples will come from a variety of applications, including (i) organ transplant decisions for patients and physicians (considering the survival curve estimates of accepting an organ and undergoing transplant, versus remaining on the waiting list hoping to receive a better quality organ in the future), (ii) catch-up scheduling for vaccinations, (iii) prenatal screening for Down Syndrome, (iv) scheduling of patients to receive a combination of services.
Nicholas Arnosti (Assistant Professor, Decision, Risk and Operations, Columbia Business School)
Georgia Perakis (Professor of Operations Research, Statistics and Operations Management, MIT Sloan School of Management)
Peak-End Demand Models and their Impact on Hard Promotion Planning Problems
Gad Allon (Professor of Operations, Information and Decisions, Wharton, UPenn)
Managing Service Systems in the Presence of Social Networks
Mohsen Bayati (Associate Professor of Operations, Information and Technology, Graduate School of Business, Stanford University)
Avoiding the Exploration-Exploitation Tradeoff in Personalized Decision-Making
Arash Asadpour (Assistant Professor of Information, Operations and Management Sciences, NYU, Stern School of Business)
Concise Bidding and Multidimensional Budget Constraints
Gad Allon (Professor of Operations, Information and Decisions, U Penn, The Wharton School)
Two Operations Management Problems in Criminology
Mor Armony (Professor of Information, Operations, and Management Sciences, NYU, Stern School of Business)
Pooling Queues with Work Averse Servers
Michael Steele (Professor of Statistics, Professor of Operations, Information and Decisions, UPenn, The Wharton School)
Expectations: What Makes Them Great?
Ilan Lobel (Professor of Operations Management in the Department of Information, Operations and Management Sciences, NYU, Stern School of Business)
Feature-based Dynamic Pricing
Junior Operations Conference
February 2 and 3, 2017
- John Silberholz (MIT)- “An Analytics Approach to Designing Combination Chemotherapy Regimens for Cancer”
- Hamsa Bastani (Stanford)- “Online Decision-Making Algorithms for Data-Driven Operations”
- Swati Gupta (MIT)- “Learning Combinatorial Structures”
- Ali Aouad (MIT)- “Revenue Management in Face of Choice Heterogeneity”
- Negin Golrezaei (USC)- “Real-Time Optimization of Personalized Assortments”
Noah Gans (Professor of Operations, Information and Decisions, UPenn, The Wharton School)
Parametric Forecasting and Stochastic Programming Models for Call-Center Workforce Scheduling
Brian Tomlin (Professor of Business Administration, Dartmouth University,Tuck School of Business)
Dual Co-Product Technologies: Implications for Process Development and Adoption
Assaf Zeevi (Professor of Decision, Risk and Operations, Graduate School of Business, Columbia University)
Chasing Bandits: (a bit of) Theory and (some) Applications
Dimitris Bertsimas (Massachusetts Institute of Technology)
Personalized Diabetes Management Using Electronic Medical Records
Pen Sun (Professor in Decision Sciences, Fuqua School of Business, Duke University)
Optimal Contract to induce Continued Effort
Jan Van Mieghem (Professor of Operations, Kellogg School of Management, Northwestern)
Collaboration and Multitasking in Processing Networks: Humans versus Machines
Mustafa Akan (Assistant Professor of Operations Management, The Tepper School of Business, Carnegie Mellon University)
Towards an Equitable Alloction of Organs Among End-Stage Liver Disease Patients
Moshe Kress (Professor of Operations Research, The Naval Postgraduate School)
When is Information Sufficient for Action? Search with Unreliable Yet Informative Intelligence
L. Beril Toktay (Brady Family Chairholder and the Scheller College of Business ADVANCE Professor, Georgia Institute of Technology)
Drawn from Efficient Implementation of Collective Extended Producer Responsibility Legislation and Design Incentives under Collective Extended Producer Responsibility: A Network Perspective
Vivek Farias (Associate Professor of Operations Management, The Sloan School of Management, MIT)
Optimal A-B Testing
Gerard Cachon (Fred R. Sullivan Professor of Operations, Information, and Decisions The Wharton School, University of Pennsylvania)
The Role of Surge Pricing on a Service Platform with Self-Scheduling Capacity
Joseph Milner (Professor of Operations Management, The Rotman School of Management, University of Toronto)
Dynamic Patient Scheduling for Multi-Appointment Health Care Programs
Forrest Crawford (Department of Biostatistics, Yale School of Public Health)
Link-tracing studies of hidden networks in epidemiology and public health
Junior Operations Conference
January 14 and 15, 2016
- Ken Moon (Stanford GSB) – “Randomized Markdowns and Online Monitoring”
- Ying Liu (NYU) – "Intertemporal Pricing under Minimax Regret"
- Kimon Drakopoulos (MIT) – “Control of Contagion Processes on Networks”
- Daniel Guetta (Columbia) – “Multi-Item Two Echelon Distribution Systems with Random Demands: Bounds and Effective Strategies”
- Dennis Zhang (Northwesten) – “Does Social Interaction Improve Service Quality? Field Evidence from Massive Open Online Education”
- Yaron Shaposhnik (MIT) – “Exploration vs. Exploitation: Reducing Uncertainty in Operational Problems”
- Mohammed Fazel-Zarandi (Toronto) – “Can Supply Chain Flexibility Facilitate Information Sharing?”
Margaret Brandeau (Professor of Medicine and Engineering, Stanford University)
Public Health Preparedness: Our Multi-Billion Dollar Problem/Opportunity
Abraham Seidmann (Xerox Professor of Computers and Information Systems, Electronic Commerce, and Operations Management WE Simon Graduate School of Business Administration, University of Rochester)
The Process Implications of using Telemedicine for Chronically Ill Patients: Analyzing Key Consequences for Patients and the Medical Specialists Practices
Sergei Savin (Associate Professor, University of Pennsylvania, The Wharton School)
Managing Office Revisit Intervals and Patient Panel Sizes in Primary Care
Christopher S. Tang (Carter Professor of Business Administration, UCLA Anderson School)
Economic Value of Market Information for Farmers in Developing Countries
Stephen Chick (Novartis Chaired Professor of Healthcare Management Technology and Operations Management Area, INSEAD)
A Bayesian Decision - Theoretical Model of sequential Clinical Trials with Delayed Responses
Francis de Vericourt (Associate Professor of Technology and Operations Management, INSEAD/ESMT)
Costis Maglaras (Columbia Business School)
Observational Learning and Abandonment in Congested Systems
Omar Besbes (Columbia University)
Should information collection affect operational decisions? The case of inventory management
Itai Ashlaghi (Assistant Professor of Operations Management, MIT Sloan School of Business)
Incentive Constraints in Random Matching Markets
Junior Operations Conference
February 6, 2015
- Maxime Cohen (MIT) – “When Data Analytics Meets Promotion Pricing”
- Daniela Saban (Columbia) – “Procurement Mechanisms for Differentiated Products”
- Angelo Mancini (Chicago) – “Dynamic Release Management: A Quasi-Open-Loop Approach”
- Yiangos Papanastasiou (LBS) – “Crowdsourcing Exploration”
Retsef Levi (J. Spencer Standish (1945) Professor of Management, Professor Operations Management, Sloan School of Management, MIT)
Systematic Approach to Manage Risks of Economically Motivated Food and Drug Adulteration in Supply Chains in China
Felipe Caro (Associate Professor in Decisions, Operations, and Technology Management, UCLA Anderson)
Adoption of a markdown decision support system at Zara
Terry Taylor (Milton W. Terrill Associate Professor of Business Administration, UC Berkeley)
Supplier Evasion of a Buyer's Audit: Implications for Motivating Supplier Social and Environmental Responsibility
Hyun-soo Ahn (Associate Professor of Operations and Management Science, Ross School of Business/University ofMichigan)
The role of cost modeling in competitive bid procurement
Carri W. Chan (Associate Professor, Decision, Risk and Operations, Columbia University)
Critical Care in hospitals: When to introduce a step down unit?
Laurens Debo (Associate Professor of Operations, University of Chicago, Booth School of Business)
Inferring Quality from Wait Time
Sameer Hasija (Assistant Professor of Technology and Operations, INSEAD)
Fleet Management Coordination in Decentralized Humanitarian Operations