Current projects and collaborations
Current projects and collaborations
Machine Learning and Interpretability, joint with Prof. Dr. C. Kleiber
The project focuses on claim-frequency and claim-severity modelling. The aim is to understand if black-box, tree-based, machine-learning algorithm can provide alternatives to the industry standard (e.g. generalized linear models) in the insurance sector. A focus of these projects is to determine the level of interpretability of these modern data science methods, which is of ample importance for the highly regulated insurance industry. Further, in order to test the applicability of modern data science methods, it is necessary to benchmark these interpretable models against linear models in the presence of large data. Potential methods under investigations to generate interpretable results are Shapley explanation values (based on the game theoretical Shapley value concept), partial dependence plots or accumulated local effect plots (visualizing the effects of predictor variables in black box supervised learning models).
Network analysis: connectivity measures and their application and interpretability, joint with Prof. Dr. M. Schienle
The project aims to provide intuition and insightful understanding for the connectedness measure introduced by Diebold and Yilmaz in 2014. A generalized variance decomposition was proposed by Koop, Pesaran and Potter in 1996 and extended by Pesaran and Shin and has been widely used in the literature ever since. Using variance decomposition for measuring the connections or linkages between entities is very attractive because it allows for a directional interpretation. Furthermore, the approach enables the analysis of a multivariate system across time. Nevertheless, the derivation of the variance decomposition components is complex and an intuitive understanding of the resulting values is not straightforward.
Credit Cycles and multi-equilibrium models, joint with Prof. Dr. M. Schienle
Credit and business cycles play an important role in economic research, especially for central banks and supervisors.