Jeppe Druedahl

Project title

Modeling Economic Agents as Deep Reinforcement Learners

What is your project about?

Data show that the economic behavior of firms and households is intentional and forward-looking but not fully rational. In economics, this is normally modelled by deriving behavior from mathematical optimization problems, which the agents are assumed to solve perfectly. My project instead assumes that agents solve these optimization problems imperfectly and make choices as if employing artificial intelligence in the form of “deep reinforcement learning”. This makes it possible to work with more complex models than previously – especially with many-sided inequality. Additionally, it paves the way for richer counter-factural analyses of economic policies than have hitherto been possible. Behavioral experiments in which subjects are asked to make consumption and saving decisions in different life cycle scenarios are used to validate these new behavioral assumptions.

How did you become interested in your particular field of research?

The fundamental economic problem is that of coordination. It is a big and complicated jigsaw puzzle: To implement the production and consumption plan of every individual frictionless, everyone’s actions have to fit together. No one can fully grasp this and everyone needs to have the right incentives to avoid a coordination breakdown. Economic crises are the most dramatic example of such a breakdown. Understanding this has been my focus since I started studying economics and is, to this day, still my focus.

What are the scientific challenges and perspectives in your project?

The economic problem of coordination is too easily solved if we assume that firms, households and other economic agents are completely rational and possess unlimited amounts of computational capacity. The challenge is to find a way of modelling bounded rationality that aligns with the empirical evidence and, at the same time, does not make our models unwieldy. Oftentimes, when behavioral bias and costs of gathering and handling information are introduced, the models become hard to solve and to simulate. Modelling agents’ behavior using artificial intelligence makes it possible to rectify this problem.

What is your estimate of the impact, which your project may have to society in the long term?

The goal of this project is to develop a new class of economic models that lies in between the standard model with optimizing agents and the so-called agent-based models with relatively simple algorithmic behavioral rules. It will become easier to make economic interventions and remodel the economy according to the prevailing political wants if we understand the economic dynamics of society better. The type of model that the project develops will in particular be able to take into account just how different individuals are with respect to preferences, perceptions of the workings of society and possible courses of action.

Which impact do you expect the Sapere Aude programme will have on your career as a researcher?

The grant will enable me to gather a group of researchers who work towards the same goal. The project will draw on my background in multiple branches of economics including studies using administrative registry-based data, behavioral economics, macroeconomics and numerical computational methods. The project focuses mainly on consumption and saving decisions, but the outlook is broader. The hope is that the methods, we develop will form the basis for a larger group that investigates this potential broadening and further developments of the core method.

Background and personal life

I am 36 years old and live in Østerbro, Copenhagen, with my partner. I am enthusiastic about both my research and teaching. In my teaching, I use programming to give my students an opportunity to work with models that are more realistic. I actively participate in the public debate and aim to communicate economic insights. For instance, I write for the Danish daily newspaper ”Information” and on Twitter. At home, I enjoy spending time in the kitchen, play board games and ride my racing bicycle.