Plenary Talks

The workshop will have several invited plenary speakers. All talks are on June 20, 2018.

Time Speaker
9:00 AM - 9:15 AM Opening remarks by the workshop organizers
9:15 AM - 9:45 AM Rene Caldentey, University of Chicago Booth School of Business
9:45 AM - 10:15 AM Ming Hu, University of Toronto Rotman School of Management
10:15 AM - 10:45 AM Sasa Pekec, Duke Fuqua School of Business
10:45 AM - 11:15 AM Break
11:15 AM - 11:45 AM Georgia Perakis, MIT Sloan School of Management
11:45 AM - 12:15 PM Wedad Elmaghraby, University of Maryland Smith School of Business
12:15 PM - 1:45 PM Lunch
1:45 PM - 2:15 PM Sid Banerjee, Cornell Operations Research and Information Engineering
2:15 PM - 2:45 PM Jacob Leshno, Columbia Graduate School of Business
2:45 PM - 3:15 PM Avi Goldfarb, University of Toronto Rotman School of Management
3:15 PM - 3:45 PM Break
3:45 PM - 4:15 PM Michael Luca, Harvard Business School
4:15 PM - 4:45 PM Damian Beil, University of Michigan Ross School of Business

See below for details on each talk.

On the Optimal Design of a Bipartite Matching System

Rene Caldentey, University of Chicago Booth School of Business
9:15 AM - 9:45 AM

Abstract: 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.

Joint work with Philipp Afeche (U. Toronto) and Varun Gupta (U. Chicago).

Dynamic Matching with Backlogged vs. Lost Types

Ming Hu, University of Toronto Rotman School of Management
9:45 AM - 10:15 AM

Abstract: We consider an intermediary’s problem of dynamically matching demand and supply of heterogeneous types in a periodic-review fashion. This problem applies to many emerging settings in the sharing economy and also includes many classic problems, e.g., assignment/transportation problems, as special cases. We provide conditions under which the optimal matching policy follows a priority hierarchy among possible matching pairs: if some pair of demand and supply types is not matched as much as possible, all pairs that have strictly lower priority down the hierarchy should not be matched. As a result of the priority property, the optimal matching policy boils down to a match-down-to threshold structure (instead of matching as much as possible in the greedy algorithm) when considering a specific pair of demand and supply types, along the priority hierarchy. We show how the optimal priority and thresholds can be fundamentally different for backlogged vs. lost types.

Operational Tools to Shape User Behavior in Platforms with Bilateral Ratings

Sasa Pekec, Duke Fuqua School of Business
10:15 AM - 10:45 AM

Abstract: We consider a two-sided service (e.g., ride-matching) platform with a ratings system that allows customers (riders) and service providers (drivers) to rate each other. The ratings system provides feedback to users, signals perceived quality of service, and consequently could impact performance of the platform. We develop an evolutionary game-theory model to study the impact of ratings system on the trajectories of user behavior over time. User responses to the ratings are endogenous, and the model also captures possible asymmetry of information provided to customers and to service providers. We characterize asymptotically stable equilibria of this dynamic system in terms of the model parameters, including price of service charged to consumers and wages earned by service providers. We use this characterization to analyze platform performance. Specifically, we show that the platform could utilize operational tools such as pricing and demand-supply management to steer the system towards a more desirable state. Furthermore, our analysis indicates that one of the most effective adjustments a platform can implement is prioritizing high-rating users in the matching algorithm.

Based on joint work with Y. Mai , Y. Hu, Z. Zou, B. Hu.

Spotting Influential Customers for Targeted Offers: From Social to Nonsocial

Georgia Perakis, MIT Sloan School of Management
11:15 AM - 11:45 AM

Abstract: The growing trend in online shopping in recent years has given rise to a wealth of data that was not available before and hence providing retailers new opportunities to personalize their services to individual customers. Examples of personalized services include targeted promotions, showing personalized assortments, and offering personalized prices. As a side benefit, knowledge of individual customer behavior can also help improve sales forecasting. To develop consumer targeted strategies, we first need to develop a demand forecasting model that captures “trends” between customers (or groups of consumers). Understanding how customers have the potential to influence each other (directly or indirectly) and create trends that lead to purchases of particular products is important when deciding on targeting promotions to the most “influential” customers.

Using the customers’ purchase information, we develop a personalized demand model that incorporates potential trends between groups of customers based on their transaction history. Currently, and to the best of our knowledge, there are no methods widely available in the industry that incorporate customer-to-customer trends and influences in demand modeling, especially in the absence of information beyond customer transactions. Unlike previous models, the customer demand estimation model can rely on a minimum of transaction data if, for example, social data is not available. In addition, the demand model we propose incorporates potential indirect and direct customer to customer trends in an interpretable manner. Furthermore, we test the customer demand model we propose with data from an Oracle client and the results improve the quality of prediction (WMAPE) around 5% relative to traditional demand forecasting models.

The richer personalized demand forecasting model we propose also allows us to determine how to offer targeted promotions to particular customers in order to improve profits. As a result, we also propose and examine an optimization model for targeted promotion offerings. Model we develop is scalable and hence can be solved quickly. Furthermore, when tested with data from an Oracle client, our targeted promotion optimization model shows profit improvements of the order of 5-12% relative to the client’s current practice.

Joint work with Lennart Baardman and Tamar Cohen (both from MIT ORC) and Setareh Borjian and Kiran Panchamgam (both from Oracle RGBU, Retail Global Business Unit).

A Market Design Perspective to Reduce Waste in Electronics and Fast Fashion

Wedad Elmaghraby, University of Maryland Smith School of Business
11:45 AM - 12:15 PM

Abstract: Whether it is due to an insatiable consumer appetite for new products, or the ability of firms to produce a lower cost thereby passing cost savings along to consumers in the form of lower prices, the volume of electronics and fashion waste that is generated each year is staggering. A number of firms have emerged to help manage this ‘excess’; adopting new, and not-so-new market paradigms in the process. In this talk, I will present the findings of an in-depth market analysis and propose several market design research directions in these markets.

Designing Decentralized Markets: Artificial Currencies and Collusion Resilience

Sid Banerjee, Cornell Operations Research and Information Engineering
1:45 PM - 2:15 PM

Abstract: With advancements in the technology for decentralized ledgers and contracts, there is a growing interest in their use in online marketplaces, and in particular, for non-monetary decentralized protocols for allocating intra-firm resources (e.g., computing-cluster cycles, vacation time) or public goods (vaccines, food donations among food banks). This opens up many new questions in the design of mechanisms. In this talk, I will talk about two such fundamental questions: how to make mechanisms efficient without requiring monetary transfers, and how to make them resilient to collusion between agents. I will discuss how traditional monetary mechanisms fail in these regards, and then present some simple yet far-reaching modifications of these mechanisms, which retrieve most of the efficiency of monetary mechanisms in a wide variety of settings.

Based on joint work with Artur Gorokh and Krishnamurthy Iyer at Cornell.

An Economic Analysis of the Bitcoin Payment System

Jacob Leshno, Columbia Business School
2:15 PM - 2:45 PM

Abstract: Owned by nobody and controlled by an almost immutable protocol, the Bitcoin payment system is a platform with two main constituencies: users and profit seeking miners who maintain the system’s infrastructure. The paper seeks to understand the economics of the system: How does the system raise revenue to pay for its infrastructure? How are usage fees determined? How much infrastructure is deployed? What are the implications of changing parameters in the protocol?

A simplified economic model that captures the system’s properties answers these questions. Transaction fees and infrastructure level are determined in an equilibrium of a congestion queueing game derived from the system’s limited throughput. The system eliminates dead-weight loss from monopoly, but introduces other inefficiencies and requires congestion to raise revenue and fund infrastructure. We explore the future potential of such systems and provide design suggestions.

Prediction Machines: The Simple Economics of Artificial Intelligence

Avi Goldfarb, University of Toronto Rotman School of Management
2:45 PM - 3:15 PM

Abstract: Recent excitement in artificial intelligence has been driven by advances in machine learning. In this sense, AI is a prediction technology. It uses data you have to fill in information you don’t have. These advances can be seen as a drop in the cost of prediction. The framing generates some powerful, but easy-to-understand implications. As the cost of something falls, we will do more of it. Cheap prediction means more prediction. Also, as the cost of something falls, it affects the value of other things. As machine prediction gets cheap, human prediction becomes less valuable while data and human judgment become more valuable. Business models that are constrained by uncertainty can be transformed, and organizations with an abundance of data and a good sense of judgment have an advantage. Details at

Digitizing Disclosure: The Case of Restaurant Hygiene Scores

Michael Luca, Harvard Business School
3:45 PM - 4:15 PM

Abstract: Collaborating with Yelp and the City of San Francisco, we revisit a canonical example of quality disclosure by evaluating - and helping to redesign - the posting of restaurant hygiene scores on Implementing a difference-in-differences strategy, we find that posting restaurant hygiene scores on Yelp leads to a 12% decrease in purchase intentions for restaurants with low scores (as predefined by the City) relative to those with higher scores. We then create a “hygiene alert” – a message that appears only for restaurants identified by the City as having “poor” operating conditions with “high-risk” hygiene violations (using the same low score threshold as above) – and find a further 9% decrease in purchase intentions. Moreover, the presence of an alert reduces the restaurant’s likelihood of getting a second alert. We conclude that disclosure policy should focus not only on what information to disclose, but also on how and where to design disclosure.

The Role of Feedback in Dynamic Crowdsourcing Contests: A Structural Empirical Analysis

Damian Beil, University of Michigan Ross School of Business
4:15 PM - 4:45 PM

Abstract: In this paper, we empirically examine the impact of performance feedback on the outcome of crowdsourcing contests. We develop a dynamic structural model to capture the economic processes that drive contest participants’ behavior, and estimate the model using a rich data set collected from a major online crowdsourcing design platform. The model captures key features of the crowdsourcing context, including a large participant pool, entries by new participants throughout the contest, exploitation (revision of previous submissions) and exploration (radically novel submissions) behaviors by contest incumbents, and the participants’ strategic choice among these entry, exploration, and exploitation decisions in a dynamic game. Using counter-factual simulations, we compare the outcome of crowdsourcing contests under alternative feedback disclosure policies and award levels. Our simulation results suggest that the full feedback policy (providing feedback throughout the contest) may not be optimal. The late feedback policy (providing feedback only in the second half of the contest) leads to a better overall contest outcome.

This is joint work with Zoey Jiang and Yan Huang.