Fraunhofer Institute for Industrial Mathematics ITWM
Fraunhofer Institute for Industrial Mathematics ITWM
Solvenzkapital Bühne
© freepik /Fraunhofer ITWM
Solvenzkapital Bühne

AI-Driven Risk Assessment for Insurance Companies

We explain and model your data within the framework of Solvency II with the support of Machine Learning.

© freepik /Fraunhofer ITWM

Solvency II Key Figures – Prediction and Explainability With Artificial Intelligence

Financial Mathematics and Machine Learning Support the Calculation of Solvency Capital Requirement (SCR)

Our team supports insurance companies in calculating Solvency Capital Requirement by developing mathematical models and AI methods that are used to assess risks and calculate capital requirements under Solvency II. These methods help companies to implement the latest research findings and meet regulatory requirements efficiently. Our expertise ranges from the development and extension of individual risk modules and improvements in the nested simulation problem to the explainability and prediction of general Solvency II parameters.

 

Solvency Capital Requirement Calculation: What is Solvency II All About?

The European supervisory regime Solvency II has been in force since 2016 – with the aim of preventing the insolvency of insurance companies and thus ensuring that they can fulfill their commitments even under extreme circumstances such as crises. Examples of such crises include natural disasters, stock market crashes or a high demand for health insurance benefits due to epidemics/pandemics. There are various options for calculating the solvency capital requirements: the standard formula, in which the SCR (Solvency Capital Requirement) for the modules is calculated in detail, an internal (risk) model, which should represent the insurance company as realistically as possible, and partial models, which represent a mixture of the two forms.

Erklärbarkeit
© Fraunhofer ITWM
We test whether a linear model is suitable for initial explanations: small variability of the coefficients when omitting data points signals robustness.

Valid and Fast Forecasts via Machine Learning Proxy Models

We enable companies to carry out a sensitivity analysis of solvency capital requirements in »real time«. Our research concept is based on a Machine Learning solution: a machine learning model, such as a neural network or a Kernel method, is trained on data from the company model. In a first step, the focus is on existing insurance data. Various data analysis tools recognize relevant risk factors of the own funds, the solvency capital requirements and the resulting solvency ratio.

We Model and Explain Your Solvency II Data

We build a surrogate model based on Machine Learning (ML) as a supplementary, easy-to-evaluate additional model for the existing risk model. This makes it possible to explain past changes as well as current estimates, forecasts and analyses of the future key figures of the respective company based on changes in market data and other parameters. We ensure continuous improvement by constantly and automatically learning new information. In addition, our proxy models indicate how confident they are in their estimate of the result. This allows the user to assess how much he or she can rely on the statements of the AI model.

The results are presented clearly in a dashboard. This can be accessed, for example, via a separate browser-based app, accessible from your intranet. All of the tools we have developed are available to you in this app, whereby internal company data does not have to be uploaded to a cloud.

Beispielbild einer Web-Ap
© Fraunhofer ITWM
Example image of a web app: An internal app with customised features that can be accessed via the browser.
Probabilistische KI-Modelle: Bayesian Neural Networks
© Fraunhofer ITWM
Probabilistic AI models – Bayesian Neural Networks: Using the example of a public data set for SCR calculation with an internal model.

PhD on Solvency Capital Requirement Calculation With Artificial Intelligence

We are working on numerous publications and analyses to continuously deepen our research and knowledge in this area and optimize our solutions. For example, our employee Mark-Oliver Wolf's ongoing doctoral thesis entitled »Mathematical and Machine Learning Aspects of the Solvency Capital Requirement Calculation« has been investigating precisely these mathematical and machine learning aspects in the calculation of the solvency capital requirement since the end of 2023. To validate his results, he developed his own simplified internal risk model in order to be able to benchmark our algorithms intensively before they are used by customers.

Prof. Dr. Ralf Korn is supervising the doctorate. He founded the »Financial Mathematics« department and headed it for many years. He contributes his in-depth expertise as a consultant and member of the Institute's Scientific Advisory Board. 

Support in the Development and Further Development of the Modeling of Individual Risk Models

Our many years of experience in financial mathematical modeling of insurance risks also gives us mature expertise in the implementation of submodules. For example, Simon Schnürch's excellent thesis on mortality modelling was developed here. Financial market risks and their modeling depend heavily on the selection of the underlying capital market, which is currently being examined by Philipp Mahler in his doctoral thesis .

Related Links: