Study guides / CFA L2: CFA Program Level II / Credit analysis models
Credit analysis models
CFA L2: CFA Program Level II · quick review
Measuring and pricing credit risk: expected exposure, loss given default, probability of default and the credit valuation adjustment; credit scores and ratings; returns under rating migration; structural and reduced-form models; credit spreads and their term structure; and how securitized debt is analyzed.
The rules
- Credit risk parameters: expected exposure is what's owed at a date if default happens then; loss given default (LGD) = exposure × (1 − recovery rate); the probability of default (POD) in a year is the hazard rate × the probability of surviving to that year; expected loss = POD × LGD.
- Credit valuation adjustment (CVA): the sum of the present values of expected losses, discounted at risk-free rates. Fair value of a risky bond = its value if it were risk-free − CVA. In a binomial tree, expected exposure at a date is the probability-weighted bond value across that date's nodes plus the coupon.
- Credit spread = YTM on the risky bond − YTM on a risk-free bond of the same maturity; roughly, the annual spread ≈ annual POD × LGD as a percentage of exposure. Default probabilities backed out of prices are risk-neutral; they usually exceed historical default rates because prices also carry risk, liquidity and uncertainty premiums.
- Credit scores rank individuals and small businesses; credit ratings rank companies, governments and their debt issues, with notching for seniority. Both are ordinal: they rank risk without giving default probabilities. Ratings are kept stable through the cycle, which makes them easy to use but slow to change; market spreads move first.
- Rating migration: expected return ≈ YTM + the expected price change from possible rating changes. Each new rating brings a new spread; price change ≈ −modified duration × change in spread; weight each by its transition probability. Downgrade risk usually leaves the expected return below the YTM.
- Structural models treat equity as a call option on the company's assets with a strike equal to the face value of debt; risky debt = risk-free debt − a put option on the assets. Default happens when assets fall below the debt. Strength: they explain why default occurs. Weaknesses: asset value and its volatility can't be observed, and the capital structure is simplified to a single zero-coupon debt.
- Reduced-form models treat default as a random event whose likelihood (the hazard rate) depends on company and macroeconomic variables. Strengths: inputs are observable, the hazard rate can change with the business cycle, and no assumption about the capital structure is needed. Weakness: they don't explain the economic cause of default, and their estimates rest on historical data that must be back-tested.
- Interpreting spread changes: a wider spread signals a higher probability of default, a lower expected recovery, or higher risk or liquidity premiums. The price effect ≈ −modified duration × change in spread, plus a convexity term for large moves.
- Term structure of credit spreads: shaped by the credit cycle, economic conditions, market supply, demand and liquidity, and issuer and industry factors. Investment-grade curves usually slope upward (default risk grows with time); high-yield curves are often flat or inverted (near-term default risk is high, and survivors tend to improve). Build curves from the same issuer's bonds against government or swap rates, using Z-spreads, or OAS for bonds with options.
- Securitized debt analysis centers on the collateral pool, the servicer and the structure (tranches, the payment waterfall, credit enhancement and a bankruptcy-remote vehicle) rather than an operating company's earning power. Granular, homogeneous pools of many small similar loans are analyzed statistically; non-granular, heterogeneous pools need loan-by-loan review. Whether the collateral amortizes also matters.
Formulas
- Loss given default: LGD = expected exposure × (1 − recovery rate)
- Probability of survival: PS_t = (1 − hazard rate)^t, for a constant hazard rate
- Probability of default in year t: POD_t = hazard rate × PS(t−1)
- Expected loss: EL_t = POD_t × LGD_t
- Credit valuation adjustment: CVA = Σ POD_t × LGD_t × DF_t, using risk-free discount factors
- Fair value of a risky bond: Value = value if risk-free − CVA
- Credit spread: Spread = YTM(risky) − YTM(risk-free)
- Approximate annual spread: Spread ≈ POD × LGD (as % of exposure)
- Price effect of a spread change: %ΔP ≈ −ModDur × Δspread + ½ × convexity × (Δspread)^2
- Expected return with migration: E(R) ≈ YTM + Σ p_i × (−ModDur × Δspread_i)
- Structural model: Equity = call on assets (strike = face value of debt); risky debt = risk-free debt − put on assets
Common traps
- CVA discounts expected losses at risk-free rates, not at the risky bond's yield.
- Default probability in year t is conditional on surviving to that year; don't apply the hazard rate to the full starting probability every year.
- Risk-neutral default probabilities from prices are higher than historical ones; don't treat them as forecasts.
- Ratings rank risk; they aren't default probabilities and they lag market spreads.
How it's tested
Vignettes give hazard rates, recovery rates and exposures (sometimes on a tree) and ask for CVA, fair value or the credit spread, the expected return under rating migration, or a comparison of structural and reduced-form models.
Practise it
The full lesson, worked examples and practice questions are in Tickmark, where every answer choice is explained.
Open the full lesson