Credit Tranche Losses and Importance Sampling
A senior tranche takes losses only after the portfolio crosses its attachment point. Plain Monte Carlo can spend most of its effort on paths with zero tranche loss. Importance sampling draws more bad scenarios and weights each draw by its relative probability. Comparing both the standard error and a conditional loss benchmark helps detect a sampling rule that produces frequent hits but unstable estimates.
How it works
An attachment and detachment point define which slice of portfolio loss a tranche bears. For a senior tranche the payoff is zero on most ordinary simulations. Importance sampling gives extra draws to adverse factor or count outcomes and applies the target-to-proposal likelihood ratio to each draw. The resulting distribution of weighted losses determines the estimator’s sampling error.
Worked example
The accompanying report uses 40 homogeneous borrowers. For its synthetic tranche at 5% asset correlation, a factored benchmark gives 1.32877 basis points of expected loss. A strong factor shift yields positive losses on roughly 66% of proposal draws, but its theoretical standard error at 200,000 draws is about 47 basis points. A milder shift combined with a conditional default tilt reduces that standard error to roughly 0.0042 basis points for the same inputs.
Validation and limitations
Compare estimates against exact conditional default counts and one-dimensional integration where possible. Check second moments and the effects of changing correlation and attachment. Expected loss is a single-period statistic. A traded tranche spread additionally depends on protection and premium payment dates, the pricing measure and discounting.
Further reading
- Importance Sampling for Portfolio Credit Risk
Paul Glasserman and Jingyi Li · Working-paper PDF
- All Your Hedges in One Basket
Leif Andersen, Jakob Sidenius and Susanta Basu · Publisher record