Editing AI for Risk Management in Banking (section)

== <span style="color: #FFFFFF;">Understanding</span> ==
Banking risk AI faces a tension unique to regulated industries: ML models offer superior predictive power, but regulators require interpretability, auditability, and model validation — favoring simpler, explainable approaches.

**Credit risk ML**: Traditional credit scoring uses logistic regression on FICO score, income, debt-to-income ratio, and payment history. ML models (gradient boosting) incorporating thousands of features — behavioral patterns, social network data, mobile metadata — achieve significantly higher discrimination (Gini coefficient). Lenders in emerging markets (M-Pesa-linked credit in Kenya, Ant Financial in China) use ML on alternative data (transaction frequency, phone usage patterns) to assess creditworthiness for people without credit history.

**Explainability requirements**: SR 11-7 and European Banking Authority guidelines require that model decisions be explainable — not just accurate. A loan rejection must be justifiable in terms regulators and customers can understand. SHAP values applied to gradient boosting models satisfy this: the contribution of each feature to the credit decision is quantifiable and auditable. This has driven adoption of "glass-box" approaches: EBMs (Explainable Boosting Machines), logistic regression with engineered features, and SHAP-explained GBMs.

**Market risk real-time monitoring**: Traditional market risk uses yesterday's VaR — a lagging measure. ML enhances this: neural networks predict intraday volatility, regime detection models flag when correlations shift (a key precursor to portfolio crisis), and graph neural networks model contagion risk between connected institutions. During the 2020 COVID crisis, correlation matrices shifted dramatically overnight — a pattern ML regime detectors could identify faster than traditional models.

**Systemic risk**: The 2008 financial crisis showed that individual bank risk models missed systemic risk — interconnected failures cascading across the financial system. Network analysis + ML maps financial institution interconnectedness (through interbank lending, derivatives, common asset exposure) to identify systemically important institutions and potential contagion pathways.
</div>

<div style="background-color: #8B0000; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">