Marketplace Algorithms and Design seminar

Organizers: Gagan Goel, Yash Kanoria, and Daniela Saban Student organizer: Judy Gan

This interdisciplinary seminar aims to cover some of the latest developments in designing and optimizing marketplaces and online platforms, leveraging data and experiments as well as tools from operations research, economics, computer science and other fields. Each talk is accompanied by two or three “guests”: topic experts who offer additional questions, perspectives and discussion points.

To receive email announcements of seminars, including Zoom information, please subscribe to our mailing list. We also live stream on Youtube.

Past Talks:

Tuesday June 30, 11am - 12pm PT. R. Preston McAfee (Google). "Ask me anything". Joined by guests Vincent Conitzer, Nicole Immorlica and Ramesh Johari.

The video recording can be found here.

I've had the fortunate experience to work on a lot of different topics -- pricing, auctions, antitrust, business strategy, market design, teach at leading engineering schools like Caltech, lead research groups at Yahoo, Google and Microsoft. Two general interest things you might want to read to ask me anything:

AI Superpowers: China, Silicon Valley and the New World Order by Kai-Fu Lee, reviewed for Business Economics, 54(3), July 2019, 185-190.

https://www.richmondfed.org/publications/research/econ_focus/2018/q4/interview

Thursday July 9, 11am - 12pm PT. Navdeep Sahni (Stanford University). "Search Advertising and Information Discovery: Are Consumers Averse to Sponsored Messages?". Joined by guest Omar Besbes.

Abstract: We analyze a large-scale field experiment conducted on a US search engine in which 3.3 million users were randomized into seeing more, or less prominent advertising. Our data rejects that users are, overall, averse to search advertising targeted to them. At the margin, users prefer the search engine with higher level of advertising. The usage of the search engine (in terms of number of searches, and number of sessions) is higher among users who see higher levels of advertising, relative to the control group. The increase in usage is larger for users on the margin who, in the past, typed a competing search engine's name in the search query and navigated away from our focal search engine. On the supply side, higher level of advertising increases traffic to newer websites. Consumers also respond more positively to advertising when local businesses in their state create new websites.

Taken together, patterns in our data are consistent with an equilibrium in which advertising compensates for important information gaps in organic listings: it conveys relevant new information, which is unknown to the search engine, and therefore missed by the organic listings algorithm. Viewing search ads, at the margin we study, makes consumers better off on average.

Thursday July 23, 11am - 12pm PT. Ramesh Johari (Stanford University). "Experimental Design in Two-Sided Platforms: An Analysis of Bias". Joined by guests Negin Golrezaei and Dave Holtz.

The video recording can be found here.

Abstract: We develop an analytical framework to study experimental design in two-sided platforms. In the settings we consider, customers rent listings; rented listings are occupied for some amount of time, then become available. Platforms typically use two common designs to study interventions in such settings: customer-side randomization (CR), and listing-side randomization (LR), along with associated estimators. We develop a stochastic model and associated mean field limit to capture dynamics in such systems, and use our model to investigate how performance of these estimators is affected by interference effects between listings and between customers. Good experimental design depends on market balance: we show that in highly demand-constrained markets, CR is unbiased, while LR is biased; conversely, in highly supply-constrained markets, LR is unbiased, while CR is biased. We also study a design based on two-sided randomization (TSR) where both customers and listings are randomized to treatment and control, and show that appropriate choices of such designs can be unbiased in both extremes of market balance, and also yield low bias in intermediate regimes of market balance.

Thursday July 30, 11am - 12pm PT. Sera Linardi (University of Pittsburgh). "Worker-Firm Relational Contracts at the Time of Shutdowns: Experimental Evidence". Joined by guest Itai Ashlagi.

The video recording can be found here.

Abstract: Exogeneous disruptions in labor demand have become more frequent in recent times. The coronavirus pandemic has resulted in millions of workers being repeatedly rehired and laid off according to local public health conditions. This may be bad news for market efficiency. Typical employment relations – which resemble non-enforceable (implicit) contracts – rely on reciprocity (Brown, Falk, and Fehr, 2004), and hence could be harmed when workers’ efforts no longer guarantee reemployment in the next period. In this paper we extend the BFF gift exchange paradigm to include a per-period probability (0%, 10%, 50%) of publicly observable “shutdown”, where a specific firm cannot contract with any workers for several periods. A Perfect Bayesian Equilibrium exists in which these shutdowns destabilize relationships, but do not harm efficiency. Laid-off workers – unemployed not due to performance - are "poached" by unmatched firms, requiring incumbent firms to rehire anew upon reopening. Workers do not shirk more in response since there is still the threat of exposure. Our experiment shows that, remarkably, market efficiency can be maintained even with very frequent stochastic shutdowns. Firms and workers adapt, changing the dynamic of relational contracts. However, the burden of maintaining productivity falls more heavily upon workers and worsens worker-side inequality.

Thursday August 6, 11am - 12pm PT. Peng Shi (USC Marshall). "Efficient Matchmaking in Assignment Games with Application to Online Platforms.". Joined by guest John Horton.

The video recording can be found here.

Abstract: An assignment game is a model of two-sided markets with transfer payments, and can be used to study the matching between heterogeneous customers and service providers with endogenous determination of prices. We study how a market intermediary can best facilitate matches in an assignment game so as to achieve a good outcome while minimizing the number of bits of information agents need to communicate. Here, "good outcomes" are formalized as ε-stable, which means that no pair of agents can deviate and both benefit by more than ε. We show that an ε-stable outcome can be found using only O(log n) bits of communication per agent whenever the horizontal component of one side's preferences is easy to predict or to satisfy. ("Easy to predict" means that those preferences can be accurately predicted by the platform after users answer short questionnaires. "Easy to satisfy" means that those agents have high horizontal preferences for many potential partners.) On the other hand, matchmaking requires Ω(sqrt{n}) bits per agent when the horizontal preferences of both sides are hard to predict and hard to satisfy. We propose near-optimal sequential protocols for both kinds of preference distributions, as well as a near-optimal simultaneous protocol for markets with easy-to-satisfy preferences on both sides. The protocols yield insights for online platforms in various industries.

Thursday August 13, 11am - 12pm PT. Youyang Gu. "covid19-projections.com - A Simple Machine Learning Model to Forecast COVID-19 Deaths". Joined by guest Vivek Farias.

The video recording can be found here.

Abstract: We present covid19-projections.com, a simple COVID-19 prediction model that adds the power of machine learning on top of a classic infectious disease SEIR model to make forecasts for all 50 US states and more than 70 countries. In this talk, we will explain how we combined the two disciplines to create an effective model that can quickly adapt to new data. Unlike other models that use a multitude of different data sources, we only use past reported deaths to predict future reported deaths. We will also provide insights into what methods worked well and what methods did not, as well as tips for how a data scientist can begin building their own forecasting models. Finally, we give an outlook into where the US may be headed as we head towards fall.

Bio: Youyang Gu is an independent data scientist and the creator of covid19-projections.com. His site has been widely cited by various organizations, including the CDC, to help inform public health decision making. Prior to his COVID-19 modeling work, Youyang honed his research skills as a quantitative researcher in trading and finance. His expertise is in using machine learning to understand data, separate signal from the noise, and make accurate predictions. Youyang completed his Bachelor’s degree at the Massachusetts Institute of Technology (MIT), double majoring in Electrical Engineering & Computer Science and Mathematics. He also received his Master’s degree at MIT, completing his thesis as part of the Natural Language Processing group at the MIT Computer Science & Artificial Intelligence Laboratory. You can find him on Twitter via @youyanggu.

Thursday August 20, 11am - 12pm PT. Juliet Schor (Boston College). "Dependence and Precarity Among Gig Workers: understanding platform heterogeneity". Joined by guest Lindsey Cameron.

The video recording can be found here.

Abstract: Since its introduction in the late 2000s, there has been growing interest in sharing economy platforms. Scholars have taken two main approaches to explaining outcomes for platform work—precarity, which focuses on employment classification and insecure labor, and technological control via algorithms. Both predict that workers will have relatively common experiences. On the basis of 112 in-depth interviews with workers on seven platforms (Airbnb, TaskRabbit, Turo, Uber, Lyft, Postmates and Favor) we find heterogeneity of experiences across and within platforms. We argue that because platform labor is weakly institutionalized, worker satisfaction, autonomy, and earnings vary significantly across and within platforms, suggesting dominant interpretations are insufficient. We find that the extent to which workers are dependent on platform income to pay basic expenses rather than working for supplemental income explains the variation in outcomes, with supplemental earners being more satisfied and higher-earning. This suggests platforms are free-riding on conventional employers. We also find that platforms are hierarchically ordered, in terms of what providers can earn, conditions of work, and their ability to produce satisfied workers. Our findings suggest the need for a new analytic approach to platforms, which emphasizes labor force diversity, connections to conventional labor markets, and worker dependence.

Thursday August 27, 11am - 12pm PT. Negin Golrezaei (MIT). "Online Learning via Offline Greedy Algorithms: Applications in Market Design and Optimization". Joined by guest Ilan Lobel.

Abstract: Motivated by online decision-making in time-varying combinatorial environments, we study the problem of transforming offline algorithms to their online counterparts. We focus on offline combinatorial problems that are amenable to a constant factor approximation using a greedy algorithm that is robust to local errors. For such problems, we provide a general framework that efficiently transforms offline robust greedy algorithms to online ones using Blackwell approachability. We show that the resulting online algorithms have O(T^{1/2}) (approximate) regret under the full information setting. We further introduce a bandit extension of Blackwell approachability that we call Bandit Blackwell approachability. We leverage this notion to transform greedy robust offline algorithms into a O(T^{2/3})(approximate) regret in the bandit setting. Demonstrating the flexibility of our framework, we apply our offline-to-online transformation to several problems at the intersection of revenue management, market design, and online optimization, including product ranking optimization in online platforms, reserve price optimization in auctions, and submodular maximization. We show that our transformation, when applied to these applications, leads to new regret bounds or improves the current known bounds.

Page updated

Google Sites

Report abuse