Within this project, we develop sparse and robust multivariate models specifically for ordinal data. The goal is to address key challenges in credit risk modeling—with relevance also for fields such as medicine and psychology.

Why is this challenging and interesting?

Credit risk modeling often relies on ordinal variables like credit ratings or default indicators, which are collected for groups of borrowers. These data are multivariate, and accounting for their dependencies improves predictive accuracy. However, existing methods face limitations:

- Variable selection: There are many potential risk factors, but only a few are relevant.
- Outliers: They can compromise model quality and stability.

With multivariate, sparse, and robust approaches, we tackle these issues—specifically for ordinal variables, for which solutions are currently scarce.

What is this needed for?

The project establishes a new framework for jointly modeling ordinal data, delivering better predictions, reduced outlier effects, and optimized credit risk assessment for financial institutions and regulatory bodies.

Team: Laura Vana-Gür (Principal investigator), Peter Filzmoser

Funding: Jubiläumsfond der österreichischen National Bank

Start: 01 October 2026, duration: 48 months

Funding: One PhD student (30 hours)