VIDEO | AI is the buzzword on everyone’s tongue these days, but not all understand the nuances in its practical applications. Anastasia Varava, Research Lead at SEBx, took to the stage at PostTrade 360° Copenhagen to give an in-depth overview of the different types of AI modelling, and how each can contribute value in real-world contexts in the financial industry.

The session was sponsored by SEB.

With ChatGPT taking over headlines, language modelling is having a moment in the spotlight right now. Anastasia Varava, however, believes that the focus should be on a type of AI that the industry is already familiar with, yet often underestimates – predictive modelling.

PostTrade 360 Nordic 2024

“We have been using predictive modelling for some years to predict revenue growth, but somehow it still didn’t scale. I think this is where most of the value is going to come in the future, because it can really bring us beyond just automated tasks,” she explains.

Watch the session video here!

First, log in (or register as a user): Login
Use your LinkedIn login if you like.

Modelling risks; giving customers hyper-personalised recommendations; predicting financial crimes and cyber attacks – these were some applications of predictive modelling that Anastasia Varava listed. “It’s not enough to just observe financial crimes happening. We need to model financial systems and transaction networks in order to predict that it is going to happen and then prevent it,” she says.

Reality check

She suggested that for predictive modelling to realise its full potential, it should be coupled with synthetic data. This would help the industry bypass the issues it is currently facing with data privacy and quality. “If we could generate high quality data… that are also plausible and representative of actual data sets but don’t contain the actual information of customers, they may be much easier to, for example, move to the cloud or even be shared externally in collaborations with other institutions. If we could have good algorithms for generating this data, it would be much easier to train models.”

“When we talk about modelling complex systems such as financial systems, we quite often simply don’t have enough historical data that covers all unexpected scenarios. If we have a way to efficiently model financial systems, creating simulated environments for various scenarios of the development of different situations, then we can really create much more efficient predictive models and take it a step further with technologies such as reinforcement learning that can actually take optimal actions as well.”

• Our news posts around PostTrade 360° Copenhagen 2023, on 11–12 October, are gathered here
• To download the 24-page jubilee event magazine, click here.
• The conference info site, with detailed agenda, is here.
• For post-event access to recorded sessions, sign up here (where you can even log in easily by your LinkedIn account). 

• By the way … are we connected on LinkedIn already, among the 3,400 post-trade pros who are? Follow us here.