Foundations of Credit Risk
The vocabulary, the master equation, and the regulatory backdrop you need before anything else makes sense.
Before we get to Altman or Merton or the Vasicek IRB formula, we need shared vocabulary. This module is the foundation everything else in the track stacks on top of. It’s the boring one. Stick with it.
What is credit risk, really?
Credit risk is the risk that a counterparty doesn’t do what they promised to do — pay back a loan, settle a derivative, deliver against a forward. That’s it. Every model in this track is just a different way of putting a number on that risk.
There are four flavors worth distinguishing:
- Default risk — the counterparty stops paying entirely.
- Downgrade risk — the counterparty’s credit quality deteriorates (rating goes from A to BBB), which moves the price of their debt even if they keep paying.
- Spread risk — credit spreads widen across the market, repricing your portfolio even if individual names haven’t changed.
- Counterparty risk — for derivatives specifically, the risk that your counterparty defaults on a contract that’s in-the-money to you.
When people say “credit risk” without qualification, they usually mean default risk, but the rest matter when you’re managing a portfolio.
The master equation
Almost every credit model in this curriculum is a way of estimating one or more terms in this equation:
Where:
- PD — Probability of Default. A number between 0 and 1 (or 0% and 100%) representing how likely the borrower is to default over some horizon, usually one year.
- LGD — Loss Given Default. If they default, what fraction of the exposure do you actually lose? It’s
1 - recovery rate. For senior secured corporate debt, LGD is often around 40%. For unsecured, it can be 70% or higher. - EAD — Exposure at Default. The dollar amount you’re on the hook for at the moment of default. For a term loan, this is roughly the outstanding balance. For a revolver, it’s harder — borrowers tend to draw down lines as they approach trouble, so EAD is usually larger than current drawn balance.
Play with the EL formula until it’s intuitive:
Notice: at PD = 2%, LGD = 45%, EAD = $100k, your expected loss is $900. That’s 90 basis points. If you price the loan at, say, 5% over your funding cost, you’ve covered EL nine times over — but only on average. The years when nothing defaults you keep all 5%; the year that one borrower defaults you lose 45% of the principal at once.
Why Basel exists
Banks have a structural problem: they make money on the spread between deposits (cheap, short-term) and loans (more expensive, long-term, risky). If too many loans go bad at the same time, the bank goes under — and because banks are interconnected, one failure can cascade.
So regulators require banks to hold capital — equity that absorbs losses before depositors take a hit. The question becomes: how much capital?
That question is what the Basel framework answers, and it’s been answered three (arguably four) times:
- Basel I (1988) — simple, blunt. Fixed risk weights by asset class. A mortgage gets 50% risk weight, a corporate loan gets 100%, and so on. You needed to hold 8% of risk-weighted assets as capital. Easy to game.
- Basel II (2004) — allowed banks to use their own internal models to estimate PD, LGD, and EAD, then plug those into a regulator-defined formula (we’ll cover this in [[module-8-portfolio-credit-risk]] — it’s the Vasicek single-factor model). Three pillars: minimum capital, supervisory review, market discipline.
- Basel III (2010, post-crisis) — tightened the definition of what counts as capital (more common equity, less hybrid instruments), added liquidity ratios (LCR, NSFR), added a leverage ratio as a non-risk-weighted backstop, introduced countercyclical buffers.
- Basel IV / “Basel III finalization” (2017, phasing in through the 2020s) — limited how much banks can reduce capital by using internal models, via an “output floor.” Mostly tightens Basel III.
For most of this track, what matters is the Basel II/III formula approach: estimate PD, LGD, EAD for each exposure, plug them into the regulatory formula, get the required capital.
What’s coming next
The rest of this track is, essentially, “how do you actually estimate PD, LGD, EAD, and the correlations between them?” Different models give different answers depending on what data you have and what you’re modeling:
- [[module-2-retail-credit-scoring]] — when you have lots of small loans and lots of historical data (consumer credit), you can build a statistical scorecard.
- [[module-3-corporate-credit-analysis]] — when you have small numbers of large loans and accounting data, you use financial ratios and models like Altman Z.
- [[module-4-structural-models]] — when the firm has publicly traded equity, you can back out an implied PD from the option-pricing structure of the balance sheet (Merton).
- [[module-5-reduced-form-models]] — when you have credit spreads from bond or CDS markets, you can back out a PD from prices.
Different tools for different problems. Pick the one that matches what you can observe.
Cheat sheet
Below this section there’s a one-page PDF glossary you can print out. It covers every term in this module plus the next two. Keep it handy.
Get new posts by email
One email per new article. No spam, no upsells, unsubscribe anytime.