You are here
Content
Development of a short survey module to measure risk aversion and prudence across outcomes and time
(in collaboration with the Technical University of Hamburg)
Recent crises highlight the importance of risks with low probability and high impact. In such cases, second-order risk aversion is insufficient to explain decision-making under uncertainty. Rather, higher-order risk preferences, such as prudence, as well as risk preferences regarding the timing of events, must be considered. These preferences influence, for example, savings decisions, negotiations, or even prevention decisions. Unfortunately, determining such preferences remains complex, time-consuming, and therefore expensive. The goal of the funded project is therefore to develop and validate a cost-effective survey module for measuring these preferences in large-scale surveys. In doing so, we contribute to empirical research and evidence-based policymaking.
Applications of Fuzzy Set Theory in Statistical Hypothesis Testing and Stochastic Claims Reserving
Insurance companies face the challenge of forecasting future claims payments under conditions of high uncertainty. Traditional methods, such as the chain-ladder method, do not account for uncertainties and do not incorporate existing expert knowledge.
The goal of the project is to develop a generalized chain-ladder method with fuzzy-adjusted run-off factors. Publicly available financial reports and structured insurance data are used to extract settlement triangles, estimate empirical distributions of chain-ladder factors, and construct nonlinear membership functions that account for heterogeneity, temporal and spatial dependencies, and economic shocks. At the same time, qualitative expert knowledge is extracted from the financial reports and methodically integrated into the analysis.
The approach combines data-driven methods such as text mining and bootstrap procedures with fuzzy set theory to integrate both explicit and implicit expert knowledge. This refines the forecasts of loss reserves and makes the risk quantifiable, offering a practical and reproducible solution for actuarial science.
Modeling and Evaluation of Modern AI Methods for the Diagnosis and Prevention of Cardiovascular Diseases
Cardiovascular diseases are among the leading causes of death worldwide. Modern AI methods enable more precise analysis of coronary CT angiography (CCTA) and offer significant advantages for diagnosis and prevention. While clinical studies have already demonstrated benefits in diagnostic accuracy and a reduction in unnecessary invasive procedures, the medium- and long-term effects on costs and healthcare efficiency remain unclear.
The aim of the project is a model-based, health economic evaluation of these AI-supported methods based on the PARAMOUNT and TRANSFORM clinical trials. Using microsimulation models, effects on clinical outcomes, quality of life, and healthcare costs will be examined to create an evidence-based decision-making foundation for use in routine care.
Prognostic Risk Assessment Using Routine Data from Statutory Health Insurance for Chronic Kidney Disease – PRED(i)CKD
(in collaboration with LiKe Healthcare Research GmbH, AOK Rheinland/Hamburg, Harz University of Applied Sciences, University Hospital Bonn, the University of Bonn, and the RWI – Leibniz Institute for Economic Research)
About ten percent of people over the age of 40 in Germany suffer from chronic kidney disease (CKD). Since the disease often progresses without symptoms in its early stages, it is frequently diagnosed late. This has serious consequences for those affected and entails high costs for the healthcare system. PRED(i)CKD addresses this issue by developing an innovative prognostic score that uses artificial intelligence and health insurance data to identify high-risk patients at an early stage.
The project follows a two-step approach: First, regression models are used to identify risk factors based on health insurance data, which allow for the prediction of various disease outcomes, such as the need for dialysis, acute kidney failure, or mortality. In parallel, powerful AI-based models, known as transformers, are used to discover new risk factors for CKD. The results of both approaches are ultimately combined into a single model that calculates the prognostic score. Concurrently, a cost analysis of the care situation is conducted.
