Editing AI for Risk Management in Banking (section)

== <span style="color: #FFFFFF;">Analyzing</span> ==
{| class="wikitable"
|+ Bank Risk AI Applications
! Risk Type !! Traditional Approach !! ML Improvement !! Regulatory Maturity
|-
| Credit scoring || Logistic regression, FICO || +15-30% Gini improvement || Accepted with explainability
|-
| PD/LGD modeling (IFRS 9) || Historical cohort analysis || Better segment granularity || Regulatory review required
|-
| Market risk (intraday VaR) || Historical simulation || Regime-conditioned NN || Research → adoption
|-
| Operational risk || Loss event database || NLP anomaly detection || Early adoption
|-
| Fraud detection || Rule systems || ML: 3-5x fewer false positives || Widely deployed
|-
| Systemic risk || Network metrics || GNN contagion modeling || Research
|}

'''Failure modes''': Regulatory non-compliance — SR 11-7 requires independent validation; ML model too complex to validate. Overfitting to economic cycle — model trained in benign credit environment fails in recession. Spurious features — ML discovers proxy variables for protected characteristics (zip code as race proxy). Model concentration risk — many banks using same vendor model creates correlated risk. Interpretability vs. accuracy tradeoff — simpler models required by regulators may sacrifice meaningful predictive power.
</div>

<div style="background-color: #483D8B; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">