Senior Professor of Computational Statistics
Profile
The Senior Professorship of Computational Statistics is dedicated to the development and application of machine learning methods in financial market and labor market research and their use in business practice. Advanced algorithms are used to identify patterns and correlations in large data sets without the need for explicit model assumptions. The aim is not only to statistically improve forecasts and classifications, but also to optimize the decision-making basis for economic analyses and measures.
The courses focus on evaluating the opportunities and risks of these methods and dealing critically with them in practice.
Current research focuses on
- Portfolio selection using machine learning methods
- Estimation of extreme financial market risks
- Analysis of network effects on labor and financial markets
Teaching
- Applied Machine Learning
- Causal Impact Evaluation
- Data Science Lab
Research interests
A. Regularization strategies for high-dimensional portfolios
This focus is on developing robust portfolio strategies that deliver optimal portfolio performance even with a very large number of investment opportunities and in times of volatile financial markets.
- A Sparse Approximate Factor Estimation for High-Dimensional Covariance Matrices and Portfolio Selection, 2024, Journal of Financial Econometrics, with Maurizio Daniele and Aygul Zagidullina
- Bagged Pretested Portfolio Selection, 2023, Journal of Business and Economics Statistics, 41, 1116-1131, with Ekaterina Kazak. Listen also to a genAI generated podcast about this paper by NotebookLM from Google.
B. Forecasting extreme financial market risks
Value at Risk (VaR) and Expected Shortfall (ES) are key risk metrics used in banking regulation to measure potential financial losses. Supervisory authorities, such as in the context of Basel III, require banks to use these ratios to ensure that they have sufficient capital reserves to cushion extreme market risks and ensure financial stability.
However, empirical experience to date with VaR and ES is ambivalent. These risk metrics have failed in times of high financial market risk, just when they were most needed. The aim of this project is to optimally combine a variety of different VaR and ES forecasts in order to achieve a more robust overall forecast. The basis for this is the bagged pretesting method developed by Kazak and Pohlmeier (2023).
- Bagged Pretested Forecast Combination for Tail Risk Measures, Paper presented at the 2024 ESIF Economics and AI+ML Meeting, August 13 - 14, 2024, Cornell University
C. Network models of peer behavior in education
This project uses network models to investigate the role of peer effects in schools. The focus is on questions such as: How do high-achieving students influence the learning behavior of their peers? To what extent is stress transmitted by peers? And which teaching organization (e.g. seating arrangements, frontal vs. group teaching) can reduce the risk of stress transmission?
To this end, specially developed econometric network models are used which, among other things, precisely depict the friendship structures within the class.
- A network model of stress contagion: evidence from the vocational classroom, Empirical Research in Vocational Education and Training, Vol. 16 (1), 2024, with T. Kärner and L. Shkoza
- Heterogeneity in Network Peer Effects, with Livia Shkoza and Derya Uysal, unpublished working paper
