“It’s a hot field and interest was great. There is no other Scandinavian bank engaged in this type of AI research,” says Patrik Karlsson.
He has a Ph.D. in financial mathematics and serves as supervisor for Hanna Hultin, the bank’s first industry Ph.D. candidate in WASP (the Wallenberg AI, Autonomous Systems and Software Programme).
In collaboration with KTH Royal Institute of Technology they are working on a research project that is studying how self-learning algorithms can be used with the aid of deep learning to take the next step in algorithmic trading.
SEB is far advanced in electronic trading. However, the current model is based on algorithms that act based on pre-programmed buy and sell signals.
“We are studying how one can use self-learning systems – ‘reinforcement learning’, together with ‘generative models’ – to allow the machine itself to decide when, where and how it should trade,” Patrik Karlsson explains.
It is thus a matter of allowing AI systems themselves to determine if there are unknown patterns or trading signals that remain undetected by people.
Thus far it involves experimental research, where the algorithms initially are allowed to train themselves in a simulated environment – roughly like Google worked on developing its AlphaGo program, which in 2015 became the first software program to beat a professional player in the Asian board game Go.
“It will take some time before we will have this in production, but it is valuable for us at SEB to be able to show that we are innovative and at the forefront of development,” says Patrik. “It strengthens our image and may increase customers’ interest in letting us handle their fundamental transactions. Plus it is helping us attract new talent!”