Valuation guides

CDO tranches and nth-to-default baskets explained

What synthetic CDO tranches and nth-to-default baskets are, their role in 2008, a worked example and how they are valued with copulas.

Definition

A tranche takes a slice of the losses on a portfolio of credits. A 3-7% tranche on an index loses nothing until portfolio losses reach 3% and is wiped out at 7%. Equity tranches take the first losses and senior tranches the last. An nth-to-default basket pays when the nth name in a small basket defaults.

Tranche values depend on default correlation: higher correlation makes large losses more likely, which hurts senior tranches and helps equity tranches.

History

Synthetic CDOs using CDS appeared in the late 1990s, and standard index tranches on CDX and iTraxx in 2003. David Li's Gaussian copula (2000) became the market model, and base correlation the quoting convention. Mispriced correlation and mortgage exposures in CDOs were at the heart of the 2008 crisis.

Examples

Tranche losses

On a 3-7% tranche of a 1 billion portfolio, the tranche is 40 million. If portfolio losses reach 5%, or 50 million, the tranche has lost 20 million, half its size.

Variations

Synthetic CDO tranche
Bespoke portfolios.
Index tranche
Standard tranches on CDX and iTraxx.
Nth-to-default basket
Pays on the nth default in a basket.

Valuation methods

Defaults are linked by a one-factor copula: Gaussian, Student-t or normal inverse Gaussian. The portfolio loss distribution is computed for each date and the tranche's expected losses give its premium and protection legs. Base correlation fits a different correlation at each detachment point to match the market skew, and the large homogeneous pool (Vasicek) gives a fast approximation for large portfolios.

Models on RiskAtoms

InstrumentPricing models
Nth-to-default basket
Gaussian copula (one factor)
Student-t copula
Normal inverse Gaussian copula
Synthetic CDO tranche
Gaussian copula (one factor)
Gaussian copula with base correlation
Student-t copula
Normal inverse Gaussian copula
Large homogeneous pool (Vasicek)
Index tranche (CDX, iTraxx)
Gaussian copula (one factor)
Gaussian copula with base correlation
Student-t copula
Normal inverse Gaussian copula
Large homogeneous pool (Vasicek)