Artificial intelligence for asset allocation

Fifteen mathematicians will this year share SEK 25 million in grants within the framework of the Knut and Alice Wallenberg Foundation's mathematics programme. The programme makes it possible, among other things, for younger mathematicians to research at one of the world's foremost mathematical environments and then return to Sweden. Petter N Kolm comes to LTH as a guest professor.

– Publicerad den 25 March 2021

Petter N Kolm is interested in developing AI for portfolio optimization of securities. Image: Mostphotos.

Artificial intelligence, or AI, has already changed vast areas of society. Autonomous computer systems can now drive cars, beat humans at chess and diagnose diseases. However, in some fields, applications have been delayed because computers are not entirely reliable – even well-functioning AI systems can, in practice, be useless if they cannot be fully trusted.

This project focuses on the development of AI for quantitative finance and asset allocation, a field that has been hesitant about using AI technologies because even very small errors can lead to huge financial losses.

Two problems will be studied. In the first step, data-driven methods for finding abrupt changes to asset dynamics will be developed. The difficulty here is in selecting the few relevant variables among hundreds or thousands of others that simply add noise. The next step involves combining the data-driven method with a machine-learning method called reinforcement learning. Reinforcement learning’s biggest successes include the AlphaGo software, which defeated the world champion at Go. Go is perhaps the world’s most complex board game, with up 10365 possible board positions.

Reinforcement learning has the advantage that it can be broadly employed, thus allowing solutions for considerably more realistic problems than in traditional methods for machine learning. For example, AI-based asset allocation could be of great importance for pension funds and private insurance foundations.