The Bank of Italy has published a research paper, “Credit Risk Assessment with Stacked Machine Learning”, assessing whether machine learning and deep learning models can enhance its In-house Credit Assessment System (ICAS) for Italian non-financial corporations used in the Eurosystem collateral framework for monetary policy implementation. The paper concludes that advanced models can increase discriminative power and that eXplainable Artificial Intelligence (XAI) can help integrate these tools into analysts’ credit assessment work, although interpretability constraints limit their ability to fully replace traditional approaches. ICAS combines a statistical model (S-ICAS) with analysts’ evaluations. The analysis finds that deep learning outperforms S-ICAS, decision tree ensemble methods deliver further gains, and a stacking meta-model combining random forests, extreme gradient boosting and deep learning improves performance again. Applying XAI techniques to the meta-model’s predictions helps identify factors driving differences between machine learning and S-ICAS outputs, supporting analysts in refining their assessments within the overall credit assessment process.
Bank of Italy 2026-01-08
Bank of Italy research finds stacked machine learning with XAI can improve ICAS credit risk assessment
The Bank of Italy's research paper finds that advanced machine learning models can enhance its In-house Credit Assessment System for Italian non-financial corporations, improving discriminative power and supporting analysts' work despite interpretability constraints. The study highlights that deep learning and decision tree ensemble methods outperform traditional models, with a stacking meta-model further enhancing performance.