Subprime & the CDO Machine, 2008

How a national bet on house prices was relabelled AAA thousands of times — and what happened when the one assumption underneath it failed.

What happened

  • 2003–06 — US mortgage lending expands into borrowers with thin documentation and low equity. Lenders sell nearly every loan onward, so underwriting quality stops being their problem.
  • 2005–07 — pools of the riskiest mortgage bonds are re-packaged into CDOs, then into CDOs of CDOs. Rating agencies stamp large senior slices AAA.
  • 2006–07 — US house prices stop rising, then fall. Delinquencies climb, and the bottom tranches begin to fail.
  • 2007 — two Bear Stearns funds collapse; the commercial-paper market freezes in August. The crisis begins in funding markets, a year before the famous part.
  • September 2008 — Lehman Brothers files for bankruptcy; AIG, having sold protection on hundreds of billions of notional without posting collateral against it, is rescued days later.

The mechanism

Tranching sorts losses: equity first, then mezzanine, then senior. It works when the pool's risks are independent — and does nothing when they are one bet repeated.
AttachDetachMezzanine (3–7%)Equity (0–3%)SeniorPortfolio lossTranche loss
  • Pooling only diversifies independent risks. Thousands of subprime mortgages across many states looked diversified. They were a single wager on the national house-price trend.
  • The models used a correlation assumption of roughly 0.3; the world delivered something close to 1. At that correlation, the subordination protecting senior tranches is decoration.
  • Re-securitisation multiplied labels, not safety — a CDO of BBB tranches is still BBB risk, however senior the new slice is called.
  • Synthetic CDOs removed the natural limit: with credit default swaps, exposure to the same mortgages could be created without any mortgages, so losses exceeded the underlying market many times over.

What it teaches

  • Ask what single variable would hit the whole pool at once. If there is one, the rating on the senior tranche is a statement about the model, not the risk (see securitisation).
  • Incentives are part of the credit analysis. A lender who sells every loan has no reason to care whether it repays — which is why risk retention rules now exist.
  • Complexity that grows with each layer is a warning, not a sign of sophistication.
  • The machinery survived, chastened. Auto ABS and post-crisis CLOs came through both 2008 and 2020 with minimal senior losses. The lesson was about what was securitised, not that securitisation is inherently unsound.