Marius Puke

This webpage presents an overview of my research and scientific work.

Education

PhD in Econometrics and Statistics (Dr. oec.)
Master in Economics
Master in Finance

Research focus

Forecast Evaluation.
Predictive Modelling.

Working Paper

T. Dimitriadis and M. Puke (2026). Statistical inference for score decompositions.

Earlier versions of this paper were presented at the following conferences.

07/2024: 44th International Symposium on Forecasting, Dijon, France.
09/2023: MathSEE Symposium, Karlsruhe Institute of Technology, Germany.
09/2023: Statistische Woche, TU Dortmund, Germany.
08/2023: 5th International Conference on COMPUTATIONAL STATISTICS, University of London, U.K.
06/2023: 5th Quantitative Finance and Financial Econometrics Conference, Aix-Marseille Université, France.
12/2022: 16th Conference on Computational and Financial Econometrics (CFE), King's College, London, U.K.

Research (peer-reviewed)

T. Dimitriadis, L. Dümbgen, A. Henzi, M. Puke, J. Ziegel (2023). Honest calibration assessment for binary outcome predictions. Biometrika.

09/2022: Statistische Woche, University of Münster, Germany.
07/2022: 42nd International Symposium on Forecasting, University of Oxford, U.K.

A. Henzi, M. Puke, T. Dimitriadis, J. Ziegel (2024). A safe Hosmer-Lemeshow test. The New England Journal of Statistics in Data Science.

M. Puke and K. Schweikert (2026). Coherent forecasting of realized volatility. Journal of Forecasting.

12/2024: 18th Conference on Computational and Financial Econometrics (CFE), King's College, London, U.K.

Thesis

Robust evaluation of point and probability predictions in economics and finance. PhD Thesis, University of Hohenheim.

Read Thesis

Software

R package "calibrationband"

This package provides a robust method for assessing the calibration of probabilistic classifiers using confidence bands for monotonic functions. In addition to testing the traditional null hypothesis of perfect calibration, the confidence bands provided by the package enable inverted calibration tests. By rejecting these tests, users can conclude that their model is sufficiently well-calibrated. Moreover, the package includes highly flexible graphical tools for conducting these assessments. It's worth noting that the computation of confidence band can be quite demanding, which is why the package relies on efficient C++ implementations. The package was devloped in joint work together with Timo Dimitriadis and Alexander Henzi.

R package "SDI"

SDI is an R package for Score Decomposition Inference that quantifies the sampling uncertainty of score decompositions obtained by linear recalibration in forecast evaluation. It supports Bregman-type scores as well as scoring functions that are strictly consistent for quantiles, i.e., generalized piecewise linear scores, and provides tools for inference on score decomposition components such as miscalibrationa and discrimination. The package is currently available as a development version on GitHub.

GitHub: development version

Other Stuff

Manim Econometrics Videos

Together with Robert Jung and Domenic Franjic, we've been able to raise 10,000 EUR in an internal university competition. This collaborative project aims to create educational videos that explain complex econometric and statistical concepts in an easy-to-understand manner, using manim - a widely used Python package for producing high-quality visualizations.

Replication Code

Inflation Chili-con-Carne Index

This project was carried out during my time as student assistant at the Department of Banking and Financial Services at the University of Hoheneheim together with Jan Swiatkowski and Hans-Peter Burghof. We tracked the food prices of European supermarket chains and monitored the prices of the ingredients needed to prepare the popular dish 'chili con carne' for more than half a year. This allowed us to illustrate food price inflation, which was soaring during the first Covid year. What started as a just-for-fun project was heavily promoted by the media; including Frankfurter Allgemeine Zeitung, Bloomberg, Stuttgarter Zeitung, Berliner Zeitung, and Fokus.