Diversified European banks are less prone to financial distress

A new research paper titled “Machine Learning for Predicting Bank Stability: The Role of Income Diversification in European Banking” has found that income diversification plays a statistically significant role in enhancing the financial stability of European banks. Published in FinTech, the study, which spans 26 countries from 2000 to 2021, uses both econometric models and machine learning to examine whether non-traditional revenue streams such as fee-based services and trading activities contribute to bank solvency and risk mitigation.

Amid ongoing debates over whether banks should prioritize traditional interest income or adopt diversified revenue models, the paper delivers critical insights using the Generalized Method of Moments (GMM) and fixed-effect models for causality testing, and machine learning models like Random Forest and Support Vector Machine (SVM) for prediction. The dual-method approach highlights income diversification as both a stabilizing factor and a reliable predictive feature in banking-sector risk assessment.

How does income diversification influence bank stability in Europe?

The study finds a significant and mostly positive relationship between income diversification and bank stability. Using key metrics, Z-score (a proxy for insolvency risk), non-performing loans (NPL), and capital adequacy ratio (CAR), the authors conclude that diversification into non-interest income sources helps reduce credit risk and boosts solvency.

Specifically, the econometric model shows that an increase in non-interest income correlates with lower NPLs, indicating that diversified banks are less exposed to credit defaults. Additionally, income diversification positively influences CAR, suggesting improved solvency buffers that allow banks to withstand economic shocks. These findings validate hypotheses H1a, H1b, and H1c, which link diversification to reductions in both credit and insolvency risks.

Moreover, higher return on assets (ROA) and improved cost-efficiency (lower EFF ratios) are also found to reinforce the positive impact of diversification. Conversely, macroeconomic indicators such as GDP growth and inflation display mixed effects, underscoring the greater importance of bank-specific variables over broad economic trends in determining stability outcomes.

How accurate are machine learning models in predicting bank distress?

To complement the econometric findings, the researchers employed machine learning models to classify banks as distressed or non-distressed based on their Z-scores, using a threshold of 3 as a benchmark. Two models, Random Forest and Support Vector Machine, were trained and tested using key financial variables, including income diversification.

Initial results showed high prediction accuracy for both models, with SVM achieving a slightly better F1-score of 0.99 compared to Random Forest’s 0.98. However, neither model could initially identify any distressed banks due to class imbalance. After applying oversampling techniques, model performance improved significantly, with SVM achieving perfect recall (1.00) and Random Forest following closely behind. Cross-validation confirmed these results, demonstrating the robustness of the models in real-world scenarios.

Feature importance analysis within the Random Forest model showed income diversification (DIV) as the strongest positive predictor of financial stability, surpassing other critical features like capital adequacy, ROA, and stock market indicators. Shapley Additive Explanations (SHAP) plots corroborated this, showing that variables such as NPLs, inflation, and unemployment had strong negative associations with stability, while diversification, profitability, and operational efficiency had positive effects.

What do the findings mean for policymakers and financial institutions?

The evidence suggests that encouraging income diversification, especially into fee-based and advisory services, can serve as a strategic buffer against systemic risks. In a post-COVID financial environment where traditional interest income has become more volatile, diversified income streams help banks stabilize revenue and mitigate credit exposure.

Second, the high performance of machine learning models underscores their utility in early warning systems. Policymakers and bank supervisors can use these models to identify signs of financial distress and implement timely interventions. The study shows that models using diversification and bank-specific indicators can accurately predict stability outcomes with over 95% reliability when properly calibrated.

Third, the findings challenge the traditional assumption that size or macroeconomic context are the most significant predictors of bank health. Instead, bank-specific decisions, such as expanding into non-traditional revenue lines or managing operational efficiency, play a far greater role in determining long-term stability.

The paper also highlights that income diversification is not without its risks. In poorly regulated environments, or when diversification leads to excessive complexity, stability can be undermined. However, within the European regulatory framework, the evidence supports a constructive role for diversification, especially when accompanied by strong risk management practices and supervisory oversight.