CDO Tranches And Correlation

Pool 100 loans together and losses will land somewhere in a range, not at one fixed number. A CDO carves that range of possible losses into layers, or tranches, stacked like floors of a building: the equity tranche absorbs the first losses (say, the first 3% of the pool's notional lost), the mezzanine tranche absorbs the next slice (3% to 10%), and the senior tranche only takes a hit once losses blow past everything below it (above 10%). Equity investors get a high yield for standing at ground level; senior investors get a low yield for the top floor, protected unless the whole building is in trouble. Each layer's size is fixed by contract, but how likely losses are to reach it depends on something separate from any single loan's default probability: how correlated the defaults are with each other.

Why correlation, not just average default rate, drives tranche risk

Under a standard one-factor model (e.g., the Vasicek single-factor framework), each borrower's default is driven by a shared systemic factor $M$ plus an idiosyncratic factor, with correlation $\rho$ controlling how much of each borrower's fate is tied to that common factor:

$$X_i = \sqrt{\rho}\, M + \sqrt{1-\rho}\, Z_i,$$

where borrower $i$ defaults if $X_i$ falls below a threshold set by their individual default probability. In plain English: at $\rho = 0$, defaults are independent, and the pool's total losses cluster tightly around the average — very unlikely to be near zero, and very unlikely to be catastrophic, so both the equity tranche and the senior tranche see fairly predictable outcomes. At $\rho$ close to 1, defaults move almost in lockstep — either almost nobody defaults (great for everyone) or almost everybody does (catastrophic for everyone) — pushing probability mass toward the two extremes and away from the middle.

Worked example 1 — low correlation protects seniors, hurts equity

With low correlation ($\rho = 0.05$), simulated pool losses cluster tightly, mostly between 4% and 7% of notional, rarely exceeding 10%. The equity tranche (0–3%) is wiped out in almost every scenario, since losses almost always exceed 3% — equity holders are nearly certain to lose their whole layer even though the average default rate looks moderate. The senior tranche (above 10%) is almost never touched, since losses essentially never reach that far — senior investors are extremely safe, protected by the sheer improbability of an extreme, clustered outcome.

Worked example 2 — raising correlation moves risk from equity to senior

Now raise correlation to $\rho = 0.40$, same average default rate. Losses spread out much more widely: a meaningful share of scenarios produce very low losses (equity survives with something left), but a new, non-trivial share produce losses above 10% — scenarios that essentially never happened at low correlation. The senior tranche, nearly risk-free before, now has genuine, non-zero loss probability. Equity's expected loss can actually fall slightly, while senior's rises from near-zero to something measurable — correlation redistributes risk from equity toward senior without changing the pool's average default rate at all.

Drag $\rho$ in the scatter above from near 0 toward 1 and watch the points cluster into the two tight diagonal arms — that same clustering, applied to a pool of loans instead of two variables, is exactly what pushes total pool losses toward "almost none default" or "almost all default" as correlation rises, thinning out the middle outcomes that would otherwise land safely inside the mezzanine layer.

What this means in practice

CDO tranches are traded and hedged largely as a correlation product: a desk long the equity tranche and short a proportional amount of senior tranche can be roughly neutral to the pool's average default rate while still being sharply exposed to changes in assumed or realized correlation — a position that changed value dramatically in 2007–2008 as realized default correlation across mortgage-backed pools turned out to be far higher than pre-crisis models assumed.