Research focus

We are interested in the intersections of digital platforms, market design and algorithmic decision-making.

Three important developments...

Growing omnipresence of digital platforms

Search engines, social networks, sharing platforms, app stores, price comparison sites and other digital platforms increasingly influence our lives, both online and offline. The market capitalizations of the firms running digital platforms belong to the highest market capitalizations in the world.

Growing omnipresence of algorithmic decision-making

At the same time, we are becoming more and more aware of the omnipresence of algorithmic decision-making, particularly machine learning, in the products we buy and the services we use. To uninformed observers, applications of such algorithms might sometimes seem almost magical.

Growing importance of the market design paradigm

Probably less prominent, but nonetheless impactful, is the shift of the economics discipline towards actively shaping markets. Instead of just analyzing their rules and institutions, economists increasingly develop concrete solutions for problems in a market setting, applying the economics tool set with an engineering mindset. Economic policy-making as well as business practice are getting increasingly influenced by this paradigm.

... which overlap

These three developments interact and sometimes even reinforce each other. For instance, digital platforms change the design of markets and sometimes create entirely new ones. A common characteristic of digital platforms is their use of algorithmic decision-making, which in turn becomes increasingly important in the design of markets. The research of the Junior Research Group “Digital Market Design” focusses on such intersections.

Digital platforms change the design of existing markets and create new ones

Four examples of how digital platforms change the rules and institutions of markets:

  • Music market
    Spotify brings together music producers and consumers through a new price-setting mechanism. Consumers buy the right for unlimited music consumption at a fixed price.
  • Hotel market
    Tripadvisor changes information provision in the hotel market by introducing a comprehensive reputation and quality management system.
  • Market for adhoc short-distance transportation
    Uber changes the price-setting mechanism in the market for adhoc short-distance transportation by introducing “surge pricing”. Uber is also reshaping the “extensive margin” of the market by allowing non-professional drivers to enter the market even only for a few hours a week.
  • Paid private accommodation market
    AirBnb largely created the market for (short-term) rentals of private accommodation, offering to rent out not only entire apartments but also individual rooms of private apartments.

Changing the design of markets is at the core of the business strategy of most of these digital platforms. The firms behind these platforms increasingly employ economists to optimize these market design interventions, sometimes with unfavorable side-effects.

Algorithmic decision-making of digital platforms becomes a design feature of markets

Algorithmic decision-making is also increasingly used to automate market design components on digital platforms. Four examples:

  • Product recommendations
    Amazon and other online retailers reportedly uses machine learning algorithms to provide recommendations, which products – besides the one displayed – might be of interest for a particular shopper.
  • “Smart bidding”
    Google Ads introduced a system to automate participation in its ad auctions. A machine learning algorithm can now be engaged to dynamically optimize ad auction outcomes.
  • Mating suggestions
    Tinder and many other dating platforms use machine learning algorithms to suggest potential partners.
  • Demand forecasting
    Uber and other platforms for adhoc short-distance transportation provide machine learning-based forecasts to drivers, where and when clients might need a ride.

The usage of algorithmic components can be both beneficial and detrimental to reach the particular aims of a market design. For instance, while they can lower search and transaction costs or enable new market mechanisms, they can also mirror or even super-charge human biases or undesirable behavior. The net effect is often difficult to untangle.

Economic incentives shape algorithmic decision-making and vice versa

Algorithms are trained and used by economic actors. As algorithms become more and more ubiquitous, the incentive schemes in play come under scrutiny. Some incentive schemes lead to beneficial algorithmic outcomes, some do not. Two examples:

  • Training data collection and generation
    The cheapest way of obtaining training data is to use existing data. However, past data collection efforts might have had a different aim, for instance to cover a particular subset of a population. As a consequence, such training data might show selection bias, which could lead to biased algorithmic results. Having the right economic incentive scheme in place to obtain the best training data available, for instance fully representative data, is key to avoid such problems.
  • Predicting policy intervention outcomes
    Policy interventions, such as unemployment benefit programs, are increasingly guided by machine predictions. Ideally, this makes targeted interventions cheaper, quicker and more precise to administer. However, the reverse might be also true, if such algorithms are not performing as intended.

Algorithmic decision-making is part of socioeconomic processes. For economists, it will be crucial not only to understand these processes, but to also actively shape these processes to deliver desirable economic outcomes.