Credit Risk & Debt Markets
Module 4 in the curriculum. This module carries significant hands-on weight: co-founding Credimi (credit scoring, underwriting, anti-fraud systems) and leading card fraud engineering at Capital One. The engineering and pipeline side is well understood; the theory --- formal loss distributions, structural models, regulatory capital frameworks --- is where the gaps live.
Why this module matters
Credit risk is the oldest risk in finance and still the largest source of bank losses. The gap between “I can build a scoring model that predicts default” and “I understand why Basel requires of capital against this portfolio” is exactly the gap between ML engineering and financial economics. Closing it connects the Credimi/Capital One experience to the theoretical foundations.
Learning roadmap
| Unit | Topic | Key concepts | Status |
|---|---|---|---|
| 4.1 | Loss Distributions & Credit VaR | , unexpected loss, loss distributions, credit VaR, correlated defaults, Vasicek single-factor model | Planned |
| 4.2 | Credit Risk Models | Merton structural model, reduced-form/intensity models, CreditMetrics, KMV distance-to-default | Planned |
| 4.3 | The Debt Universe | IG vs HY, leveraged loans, mezzanine, subordination, covenants, distressed debt | Planned |
| 4.4 | Structured Products & Securitization | ABS, MBS, CDOs, CLOs, tranching, waterfalls, credit enhancement, the 2008 crisis | Planned |
| 4.5 | Basel Framework | Basel II/III/IV, risk-weighted assets (RWA), capital ratios, standardized vs IRB approaches | Planned |
Core ideas to internalize
Expected vs. unexpected loss
Expected loss is the mean of the loss distribution --- it is a cost of doing business, provisioned for through loan pricing and reserves:
Unexpected loss is the volatility around that mean --- the tail risk that capital must absorb. Credit VaR is typically defined as the -quantile of the loss distribution minus EL. The challenge: defaults are correlated (recessions hit everyone), so portfolio credit risk is not just a sum of individual risks.
Structural vs. reduced-form models
Merton (1974): a firm defaults when its asset value falls below its debt at maturity. Equity is a call option on :
This connects beautifully to options pricing theory. KMV operationalizes this by computing “distance to default” from equity prices.
Reduced-form models (Jarrow-Turnbull, Duffie-Singleton) treat default as a surprise event governed by a hazard rate , calibrated from credit spreads. They are more tractable for pricing credit derivatives but offer less structural intuition.
The Basel capital logic
Basel maps credit risk to capital requirements. Under the IRB approach, the Vasicek single-factor model drives the capital formula:
\sqrt{1-\rho}}\right) - \text{PD}\right]$$ where $\rho$ is the asset correlation parameter. This is the bridge between portfolio credit theory and regulatory practice. ## Connections to experience - **Credimi**: The scoring models estimated PD; the underwriting logic implicitly set LGD and EAD boundaries. This module adds the portfolio-level view (correlated defaults, capital allocation) that a lending platform needs at scale. - **Capital One**: Fraud losses are a special case of credit loss. The engineering of real-time decisioning maps to the "unexpected loss" framework --- you provision for baseline fraud (EL) and hold capital/reserves for spikes (UL). ## Key references - **Bluhm, Overbeck, Wagner** --- *An Introduction to Credit Risk Modeling* - **Duffie & Singleton** --- *Credit Risk* (reduced-form focus) - **Hull** --- chapters on credit risk and credit derivatives - **Basel Committee publications** --- Basel II/III/IV framework documents