Module 33 · Computing credit risk
MBS credit risk — default racing prepayment
“MBS credit risk” starts with a hard fork. Agency MBS — the Fannie, Freddie and Ginnie pass-throughs from Module 12 — carries essentially no credit risk for the investor: defaulted loans are bought out of the pool at par under the guarantee. All the risk is prepayment and convexity. Non-agency (private-label) MBS has no guarantee, and there the credit modeling is as loan-level and intensive as anywhere in fixed income.
The non-agency framework is competing risks: every month, each loan can prepay, default, or survive — and the two hazards compete, because a borrower who refinances can never subsequently default. Fast prepayments quietly de-risk the pool; a rate environment that shuts off refinancing traps weak borrowers in it. You cannot model mortgage credit without modeling prepayment at the same time.
The default side is driven above all by one number: mark-to-market LTV— the loan balance against the home's current value, rolled forward on a home-price (HPA) path. It drives both halves of expected loss:
- Default probability — borrowers with equity in the home sell rather than default; negative-equity borrowers default at multiples of the base rate. FICO, DTI, documentation and seasoning layer on top.
- Severity — what's lost when they do: the shortfall between the loan and the home's value net of foreclosure and liquidation costs, stretched by the foreclosure timeline (judicial vs non-judicial states) and servicer advances.
Because home prices sit inside both PD and severity, collateral losses are convex in the HPA path — which is why the right headline metric for non-agency paper is a loss-adjusted OAS or scenario loss under explicit HPA paths, not a single spread duration. A bumped CS01 hides most of what you care about.
One hybrid case ties this module to the last one: CRT(credit-risk transfer — Fannie's CAS, Freddie's STACR), where the GSEs sell their guarantee risk to investors as tranchednotes on a reference pool. It's mortgage credit in Module 32's clothing — with the twist that fast prepayments retire the deal early and de-lever the tranches, so prepayment speed acts as a credit variable, not just a duration one. Commercial-property credit (CMBS) has its own module: DSCR, LTV and balloon risk.
🎛 MBS credit lab
MTM LTV in 5y
89%
LTV ÷ (1+HPA)^T
Severity
21%
loss per defaulted $
Cum. defaults
7.9%
competing with prepays
Escaped via prepay
31.5%
refinanced — can't default
Collateral loss
1.65%
defaults × severity
Your credit loss
1.65%
no guarantee
Losses are convex in the home-price path: falling prices push the mark-to-market LTV above 100%, severity turns positive and grows, while rising prices drive severity to zero no matter how many loans default. Prepayment and default are competing risks — a borrower who refinances leaves the pool and can never default, so a high CPR quietly de-risks the credit. Educational tool — not investment advice.
How the lab computes it
MTM LTV = LTV / (1 + HPA)^T severity = max( 0, 1 − (1 − liq.cost) / MTM LTV ) cum defaults = CDR/(CDR+CPR) × (1 − e^−(CDR+CPR)·T) collateral loss = cum defaults × severity agency: investor loss = 0 (guarantee absorbs it)
The CDR/(CDR+CPR)factor is the competing-risks split: of the loans that leave the pool, that's the share leaving through default rather than refinancing. Production models estimate both hazards loan by loan (LTV, FICO, rate incentive, seasoning, geography) and run them through a cashflow engine over many HPA/rate paths — the lab is one loan-cohort slice of that machine.
Things to try
- • Sweep HPA from +5% to −5% with an 80 LTV — severity sits at zero, then snaps positive and grows. That kink is why mortgage credit losses are convex in home prices, and why 2007-vintage models trained on rising prices failed.
- • Hold everything fixed and raise CPR from 5% to 30% — cumulative defaults fall even though the default rate never changed. Refinancing rescues borrowers before they can default.
- • Now cut CPR to 2% (rates spiked, refi shut) with negative HPA — defaults and severity rise together. This is the trapped-borrower scenario, and it's the one that hurts.
- • Flip to agency with the same ugly assumptions — your credit loss reads zero while the collateral loss doesn't change. That difference is exactly what CRT investors are paid to absorb.