Module 32 · Computing credit risk

Structured credit risk — losses through a waterfall

For a corporate bond, credit risk is a spread question. For a tranche of a CLO, ABS or CMBS deal, it's a distribution question, computed in two stages: first model how much the collateral pool loses, then push that loss through the waterfall to see how much reaches your tranche.

Stage one is a handful of collateral assumptions: CDR (the annual default rate), severity(loss per defaulted dollar), and how long the pool is outstanding. Together they give a cumulative collateral loss — defaults × severity.

Stage two is pure structure. Your tranche has an attachment point(the subordination beneath you — losses up to here are someone else's problem) and a detachment point(past here, you're wiped out). The payoff is a hockey stick: 0% written down below attach, 100% above detach, and a violently steep ramp in between. Three numbers summarize your position on it:

  • Credit enhancement / cushion — the distance between current collateral losses and your attachment point. This is the state variable to track deal by deal, month by month.
  • Break-even CDR — the constant default rate at which losses just reach your attachment. The margin of safety expressed in the collateral's own units.
  • Structural leverage — inside the loss window, each 1% of extra collateral loss destroys 1/thickness of your tranche. A 3%-thick mezz bond loses 33% of face per point of collateral loss.

That leverage is why tranche risk is non-monotonic and non-linear in ways single-name credit never is: one large obligor default can wipe a thin mezzanine tranche while barely denting the senior above it, and equity and mezzanine tranches take opposite sides of default correlation (covered in Module 29).

🎛 Tranche credit lab

Pick a tranche:
3%
40%
5y
10%
5%

Tranche

1015%

attach – detach

Cum. defaults 5y

14.1%

1 − (1 − CDR)^T

Collateral loss

5.7%

defaults × severity

Loss cushion

4.3%

attach − collateral loss

Tranche writedown

0%

of tranche face

Break-even CDR

5.59%

first dollar of loss

Wipe-out CDR

8.97%

tranche fully written down

The tranche payoff is the hockey stick above: nothing until collateral losses reach the attachment (amber line), then total destruction by the detachment (red line). Tranche credit risk is a question about the collateral loss distribution relative to those two lines — not a spread-duration number. Educational tool — not investment advice.

How the lab computes it

cum defaults    = 1 − (1 − CDR)^T collateral loss = cum defaults × severity tranche loss    = clamp( (col.loss − attach) / thickness, 0, 1 ) break-even CDR  : solve  collateral loss(CDR) = attach leverage        = 1 / thickness   (inside the loss window)

Real deal engines (Intex-style) run full cashflow waterfalls — coverage tests, excess spread, sequential vs pro-rata paydown — and desks run whole vectors of CDR/CPR/severity scenarios rather than single points. But every one of those scenarios lands back on this picture: where did collateral losses end up relative to attach and detach?

The portfolio trap: obligor overlap

The credit enhancement math above is per-deal. The risk that actually bites a structured-credit portfolio is that the same borrowers appear in many deals — a leveraged-loan issuer might sit in dozens of CLOs at once. Deal-level aggregation makes ten CLO positions look diversified when they share most of their collateral. Proper risk requires look-through to the obligor: map every underlying borrower across every deal, aggregate the exposure, and net it against any single-name or index hedges. (And on capital: securitizations get standardized regulatory treatment — SEC-ERBA/SEC-SA under FRTB — no matter how good your internal model is.)

Things to try

  • • Pick the Mezzanine preset and raise CDR slowly — nothing, nothing, nothing, then the tranche shreds over a ~1% CDR range. That cliff is the defining feature of mezz risk.
  • • Compare break-even CDR across the three presets at 40% severity — senior survives default rates that vaporize the equity several times over. Same pool, different bonds.
  • • Halve the thickness of the mezz tranche — break-even CDR barely moves, but the leverage inside the loss window doubles. Thin tranches don't drift; they gap.
  • • Drop severity from 40% to 20% (senior-secured loans vs subordinated collateral) — every break-even doubles. Recovery assumptions matter as much as default assumptions.