Markets are getting more and more complex. For the market designer, creating optimal mechanisms for these markets “by hand” is getting harder or even impossible. And for the market participants, evaluating the wealth of possible choices in these markets is getting infeasible, so they typically have to settle for sub-optimal decisions.
Sven Seuken’s research group will combine techniques from machine learning with market design to develop new market mechanisms that are more efficient, less manipulable, fairer, and easier to use for market participants. In the future, this new approach can be used by other researchers from Artificial Intelligence, Game Theory, and Economics to design new market mechanisms.
Society at large will directly benefit from the new approach, because Seuken’s group will tackle two concrete market design applications within the scope of the ERC project. Specifically, they will address: spectrum auctions and school choice mechanisms.
Spectrum auctions: Every few years (and again in 2019), the Swiss Government uses a so-called “spectrum auction” to sell licenses for radio spectrum (3G, 4G, 5G) to mobile network operators such as Swisscom, Salt, and Sunrise. Designing these kinds of auctions is very challenging because many different but related licenses are sold at the same time. Seuken’s research has already shown that current auction designs have multiple shortcomings, potentially leading to inefficiencies and revenue shortfalls on the order of 100s of millions of Francs per auction.
The ERC project will enable the design of machine learning-based spectrum auctions that are less manipulable, fairer, more efficient, and more profitable. This will lead to better phone networks and a more competitive marketplace. Thus, in the long-term, end-users will benefit from a better quality of service at lower prices.
School choice: Every year, millions of students around the world enter a new public school. But schools have limited capacities and some schools may be more popular than others such that not all students can go to their most preferred school. “School choice mechanisms” (as used in parts of the USA, Germany, Mexico, China, etc.) assign students to schools based on a ranking of schools submitted by each student. However, even exploring all possible choices may be infeasible for the students, as there may be hundreds of possible schools in large cities like New York City.
The ERC project will enable the design of new school choice mechanisms that use machine learning to help students navigate their complex set of choices during the application process. Seuken will study whether such a machine learning-supported design can lead to a more efficient assignment of students to schools, thereby improving students’ educational options.