Marcel graduated from the University of Cologne in 2020 with a MSc in Mathematics, after completing a BSc degree in Mathematical Economics at the University of Mannheim in 2018 during which he also spent one semester at the University of Florida in 2016. While his research interests in Mannheim solely revolved around the theory of stochastic processes, a subsequent internship at a risk-consultancy firm sparked his fascination of financial mathematics as well as time-series analysis and machine learning. Through his supportive role in concepting and implementing a prediction algorithm of EEX spot market prices for a multinational energy company, he gained crucial modelling expertise for his minor in Informatics in Cologne. There, his dissertation on typical distances in scale-free random networks supported a research group in their analysis of spatial networks. Outside of his studies, Marcel is a passionate tennis player who collects scuba diving and kickboxing trainer licenses.

Fun fact(s):
I am a certified kickboxing and scuba diving instructor.

Research project title:
Robust Adaptive Dynamic Programming for the Newsvendor Problem

Supervisor(s):
Alex Cox, Clarice Poon

Project description:
In data-driven stochastic optimisation problems that account for model uncertainty, the decision maker can either act myopically based on the knowledge that has been acquired so far or exploratory by choosing a sub-optimal action that may gather more information about the system to improve future decisions. This exploration-exploitation trade-off is most prominently encountered in the newsvendor problem where a retailer can only observe the number of products sold but not the excess demand, and is encouraged to learn about the unknown demand distribution to make more informed inventory decisions. Marcel’s work focuses on approximating the optimal solution in this planning problem in such a way that the decision maker can learn the true characteristics of the demand robustly under model misspecifications. Based on concepts from optimal transport theory he plans to develop a method aimed at adaptively reducing the infinite space of attainable future models to a finite number of distinct forms; which allows for an effective computation of the intractable Bellman equation. He intends to extend this method to a broader range of uncertain stochastic optimisation problems which may also allow for a scenario reduction technique in algorithm planning for Reinforcement Learning applications.