A new ECB paper shows how random forest models can improve default probability estimates for European companies.

A working paper from the European Central Bank examines random forest models for estimating the probability of default of European non-financial companies. It draws on balance sheet data from Orbis and the Anacredit bank credit register and assesses the implications for banking sector stress tests.

The European Central Bank has published a working paper examining random forest models for estimating the probability of default of European non-financial corporations. The data comes from balance sheet records in the Orbis database and the Anacredit credit registry maintained by banks. The study finds that random forest models show stronger risk sensitivity than classic logistic regression and reveal a non-linear relationship between scenario severity and default probability.

For banks and supervisors, the study offers concrete pointers on how machine learning models could complement or reshape regulatory stress tests going forward. The authors also show how random forest based default probabilities can be used within a network of banks and firms to trace how adverse scenarios spread through loan exposures as a key transmission channel. This produces a more granular view of individual banks risk indicators.

It remains open when and how supervisors will adopt such models into official stress test procedures, since explainability and regulatory approval still face higher hurdles for machine learning approaches than for classic models. Banks with their own risk departments can already test the methodology, but a binding supervisory standard has not yet been set. The study offers a research foundation rather than an immediate call to action.

This story was produced automatically from the source named above and checked by software before publication. The image is symbolic and shows neither the event nor a real person. How this paper is made