Less Data, More Valid Forecasts

Solvency Capital Calculation With Artificial Intelligence

Interview / September 07, 2023

For a very long time now, our »Financial Mathematics« department has been working on determining solvency capital, i.e. the capital that insurance companies must hold in reserve in order to be solvent at the end of the year. In order to be able to calculate the so-called solvency ratio, the researchers also rely on artificial intelligence. Prof. Dr. Ralf Korn explains how research in this area has developed. He has headed the »Financial Mathematics« department for many years and now contributes his expertise as a consultant and member of the Institute's Scientific Advisory Board.

How can Artificial Intelligence protect insurance companies from risks that are difficult to calculate?

There is no protection, of course, but our methods help to make better decisions. We offer insurance companies a neural network that trains with existing data and the company's internal model. The goal is a sensitivity analysis of the solvency capital in »real time«. We now integrate external data into this internal model.

You have to explain this to us.

What is important in the solvency capital calculation: There is not only the insurance company, there is also the world! And it must flow into the sensitivity analysis, in the form of external scenarios. We integrate uncertainties into our calculations in order to be able to go further in our predictions. To do this, we don't have to know the exact state of today, but can also look at what would have happened if we had perhaps made a small mistake. We simulate the state of the world at certain points and then evaluate the insurance stocks at exactly those points. And because this evaluation is very time-consuming and one would have to simulate about 10,000 states to know about the world, one has not only an evaluation problem but also a time problem: An evaluation of the assets and liabilities of the insurance company in one state of the world at a time takes about one minute.

And this is where Artificial Intelligence comes into play?

Yes, because we can simulate 10,000 states of the world in real time, so to speak, i.e. the development of interest rates, mortality, stock markets and so on. What takes time is valuing insurers' liabilities and assets internally. So we take an approach where we actually only do a very rough valuation by rolling the dice, as it were, only once in each state of the world. It balances out because all these 10,000 valuations are very rough approximations, but if you do a regression calculation – a balancing calculation – you get an amazingly good result. And that's already after a few hours, because we calculate in parallel on a very large cluster.

What’s next?

Currently, we are researching whether you really need all 10,000 valuations for a valid forecast and have created a demonstrator with which we can already show things well. Fortunately, we have a new colleague, Mark-Oliver Wolf, who is doing research in the field of »Valuation Models and Algorithms for Structured Products and Derivatives« using machine learning methods.

Together with the statistician Gerhard Stahl (former Chief Risk Officer of HDI-Talanx), I have developed a revolutionary model, which unfortunately cannot make much money from an ITWM point of view, but which can possibly be used to invest a lot of money in wind farms or other sustainable investments instead of just letting it lie around.

We use a reduced model for risk assessment, in which the fluctuation range of the insurance company's own funds is used as the forecast basis. This is an established method in equity port folios. Here, we look at the development over 28 quarters, so we have much less data. Nevertheless, the forecast quality is very good and enables insurance companies to invest more and still meet the solvency ratio.

Prof. Dr. Ralf Korn in conversation — © Fraunhofer ITWM
Prof. Dr. Ralf Korn has established our department »Financial Mathematics« and continues to support the Fraunhofer ITWM as a consultant.

Interview

Less Data, More Valid Forecasts

Solvency Capital Calculation With Artificial Intelligence

How can Artificial Intelligence protect insurance companies from risks that are difficult to calculate?

You have to explain this to us.

Mehr Informationen

And this is where Artificial Intelligence comes into play?

What’s next?

Contact Press / Media

Prof. Dr. Ralf Korn

Contact Press / Media

Ilka Blauth