The aim of this work is to introduce an innovative methodology for performing risk attribution within a multifactor risk framework. We applied this analysis to the assessment of systemic, climate, and geopolitical risks relative to a representative sample of Eurozone banks between 2011 and 2022. Comparing the results to the output of a bivariate approach, we found that contemporaneous tail crises generate combined equity losses exceeding partial analysis estimates. We then attributed the combined risk to each factor and to the effect of their interaction by employing our proposed frequency-based approach. For our computations, we used multivariate GARCH, Monte Carlo simulations, and a suite of Eurozone-specific factors. Our results show that total combined risk is on average 18% higher than traditional systemic risk estimates, that climate risk more than doubled in our period of analysis, and that geopolitical risk surged to over 5% of total combined risk. Our climate risk estimate is in line with the results of the 2022 European Central Bank climate stress test, and our geopolitical risk measure shows a positive correlation with the GPRD and Threats index.
Provenance. PONDUS-AI was developed by the author from early 2026 through three versioned releases (see Section 8), under the technical identifier fin-ai-governance. It was integrated as a case study in a European Innovation Council Pathfinder proposal and is published here to establish a public, timestamped record of authorship prior to resubmission.
We investigate the effectiveness of three different climate metrics in identifying green banks within a sample of large Eurozone and US lenders in the February 2019 – April 2022 period. We compute weight modifiers determined by relative exposure to stranded assets, environmental ratings and Scope 2 emissions and apply them to create sectoral portfolios robust to climate events. The adjusted portfolios’ risk-return performance is compared against a market capitalization benchmark and an optimized portfolio. The results show that the selected climate loss proxy overperforms in the Eurozone, succeeding in creating an effective climate tilt while containing active risk. Both emission-adjusted and rating-modified portfolios work as well, albeit less effectively. Conversely, the results with respect to US banks are inconclusive, with no metric consistently overperforming, likely due to the absence of a homogeneous Scope 3 indirect emissions reporting standard and differences in the regulatory frameworks, central banks policies and accounting practices.