Generic framework for a coherent integration of experience and exposure rating in reinsurance

Hello and welcome. It's September 30th, 2024. We're in Active Inference Guest Stream 89.1, Generic Framework for a Coherent Integration of Experience and Exposure Rating in Reinsurance with Stefan Werninger and also joined by Roland Borghi. So thank you both for joining. Stefan, to you for a presentation and we'll look forward to that and the discussion. Okay, thank you, Daniel. Welcome everyone. Let me quickly give an overview of what I'm going to present. It's about reinsurance and reinsurance, maybe some of the people on the call are familiar with what reinsurance is. If you are not, no problem. I'm going to explain everything you need to know. Basically what reinsurers do, they provide coverage to insurance companies and when they provide the contract to cover the risks, they of course need somehow to rate this risk. And there's two different approaches which are traditionally used. One is experience rating, the other exposure rating. And what I'm going to present is a new framework which is integrating the two into one integrated framework basically. And just to what you can see here, what is highlighted, there are three components to it. The first component is a generative model. The second is, I mean, the generative model, he will create artificial or simulated claims. And what we need to do before we can do an analysis, we need somehow to reduce this high dimensional complex objects into some smaller number of what I call reduced random variables. And then we use invasion inference to basically calibrate the model. This is basically what it's supposed to do. What you see on the bottom, I made a reference to some people in the history. So the traditional approach where you use experience rating, you're taking the claims from the past, you somehow project this into the future and use this data as a model for the future claims activity. So that's why I put there Kepler, who basically took the data from Tycho Brahe and then he derived his laws for the motion of the planets. And then a contrast is Newton, which basically can derive these laws by some much more fundamental principles. So basically in one case, you model the data and the other case, you model the world. So this is the difference between the two. Of course, in biology, we would have the Aristotle on the one hand side and then Darwin on the other side and from the statistics, what you see on the left side, this is the father of the classical statistics of Fisher and on the other side, they put the base. So this is where the frame where we are operating. Now to my talk, I will first give some background information and explain why this is relevant. Then I will somehow make a link to active inference, then also point out to some of the challenges which we are facing in this attempt. And then I'll provide with a roadmap for a solution. And then I will show some examples. And just for your information, I will not show any formulas here. So whenever you want to look into the formulas, we can go back to the paper, you find all the technical details. So I'll just describe with the help of numbers and some charts or some graphics, what I'm going to or the basic ideas are. So just if you expect any formulas, you will be disappointed. Okay, so first, what I've done, I've taken some real numbers from a re-intruance company. And this is Swiss Re. And from Swiss Re, I've taken not the whole business, but just one segment, which is the PNC reinsurance, which is property, casualty and specialty lines. Two other segments, which this re-hears is a corporate solution. So this is then direct insurance provided to large corporations. And there's also life and health. So what I'm showing here is only PNC reinsurance, so contracts with insurance companies. And Swiss Re has developed about 20 or 25 years ago, an internal framework for economic valuation of the business. And it has started in 2013 to publish the numbers. So EVM results, which is an economic valuation of the business…