Lecture: Market Design and Machine Learning (Spring 2019)

Lecturer: Prof. Dr. Sven Seuken
Teaching Assistant: Ludwig Dierks
Teaching Language English
Level MSc, PhD
Academic Semester Spring 2019
Time and Location Wednesday, 10:15-12:00, BIN-2.A.01
AP (ECTS): 6 (including a mark)
Office Hours Prof. Dr. Sven Seuken: email for appointments, BIN-2.B.02


Course Content

Over the last two decades, the field of market design has developed sophisticated techniques to design practical market mechanisms with good economic and computational properties (taking into that market participants are strategic). At the same time, the field of machine learning has developed more and more powerful techniques to generalize from data, adapt to changing environments, and thereby improve a system’s performance with experience. This course explores how these two seemingly unrelated fields can be usefully combined. In particular, we will discuss how we can use machine learning techniques to design better market mechanisms (like auctions or matching mechanisms) and how we can incorporate machine learning algorithms into the operation of complex marketplaces (like Uber, eBay, or AirBnB) to improve their performance. Students will read key papers from the literature (theoretical and applied) and get hands-on experience by working on a project combining market design with machine learning.

Teaching Format and Setup

The course has two parts. The first part is structured like a PhD-level seminar, where students read a paper each week, write a brief response essay (0.5 pages), and most of the time in class is used for interactive discussions. Each week, one or two students present the week's paper and lead the discussion.

The second part of the course consists of a project that combines market design and machine learning (on which students can work alone or in teams of 2 or 3 students). The projects can extend existing work that combines ML and market design, or they can find news ways of combining ML with market design (e.g., in novel domains). The projects can be theoretical, empirical (using data), or experimental (using simulations). The students are free to choose their own projects. Successful projects will ideally lead to a workshop or conference paper submission.



  1. Introduction [Introduction Slides] (PDF, 317 KB)
  2. Marketplaces, Markets, and Market Design (Roth, 2018) |  [paper]
  3. Fast Iterative Combinatorial Auctions via Bayesian Learning (Brero et al., 2019) |  [paper]
  4. Machine Learning-powered Iterative Combinatorial Auctions (Brero et al., 2018) |  [paper]
  5. Payment Rules through Discriminant-Based Classifiers (Dütting et al., 2015) |  [paper]
  6. Optimal Auctions through Deep Learning (Dütting et al., 2018) |  [paper]
  7. Project Idea Presentation
  8. Improving refugee integration through data-driven algorithmic assignment (Bansak et al., 2018) |  [paper][supplementary]
  9. A Simulation Approach to Designing Digital Matching Platforms  (Fradkin, 2019) | [paper]
  10. No class (time to work on projects)
  11. No class (time to work on projects)
  12. Project: Final Presentations

Further reading:

  1. Reducing Mechanism Design to Algorithm Design via Machine Learning (Balcan et al., 2007) |  [paper]
  2. Canary in the e-Commerce Coal Mine: Detecting and Predicting Poor Experiences Using Buyer-to-Seller Messages (Masterov et al., 2015) | [paper]
  3. Search, Matching, and the Role of Digital Marketplace Design in Enabling Trade: Evidence from Airbnb (Fradkin, 2018) |  [paper]
  4. Adaptive-Price Combinatorial Auctions (Lahaie and Lubin, 2018) | [paper]
  5. Matching Systems for Refugees (Jones and Teytelboym, 2018) | [paper]


This course requires prior knowledge in (1) market design/mechanism design and (2) machine learning. To obtain the prior knowledge for market design, the successful completion of a course covering basic topics on market design (such as auction theory, mechanism design, matching, etc.) is required. Courses with the necessary background include “Economics and Computation” and “Introduction to Market Design” at UZH, as well as “Algorithmic Game Theory” at ETH. To obtain the prior knowledge for machine learning, any introductory course on machine learning is sufficient. Students who have not taken such courses beforehand may be eligible but must contact the instructor ahead of time to request explicit  consent.

Target Audience

Recommended for MSc and PhD students.

Learning Objectives

  1. Understand how machine learning can be useful in the design of specific market mechanisms and in the design of complex marketplaces
  2. Understand the difficulties involved when combining machine learning with market design techniques.
  3. Be able to read advanced research papers.
  4. Be able to critically reflect on and discuss a advanced research papers.
  5. Be able to identify how machine learning could help solve a new market design problem.
  6. Successfully complete a project combining machine learning and market design.


Examination and Grading

  • Presentation of papers and leading class discussion: 20%
  • Response essays and class participation: 30%
  • Project: 50%