Random recovery model would increase tranche liquidity, says Merrill Lynch

Given that changing the model would change the deltas in a roughly equivalent way for both bespoke tranches and index tranches, I do not think that that there would be any direct influence on index tranche prices. What it will affect is the single name hedges which may need substantially rebalancing though which way this will pull prices depends on the inventory that needs re-hedging.

What should be the consequences if banks start using this model for hedging their bespoke books ? The AH model implies lower SS deltas and higher mezz deltas -> What will be the consequences for index tranche prices ?

In view of the continuing widening of senior IG tranche spreads, a very timely assessment of what is an intuitive enhancement of the standard base correlation framework. Independent implementations of this model such as CDO2, should make calibration discrepancies easier to reconcile.

This model is inconsistent. Adding stochastic recovery to a model which is not arbitrage free, lacks dynamics, and has no clear economic or financial motivation is simply a way to improve the fit by adding another parameter. Those who are interested in a bottom-up dynamic arbitrage-free stochastic recovery model where recovery is correlated with default and all maturities are treated simultaneously might want to talk to Julius Finance.

This is one of the first new models that we have seen broadly taken up by the larger banks. It does not solve every problem but is a very attractive incremental evolution of the standard gaussian copula model that addresses problems with calibration, negative deltas, and super senior pricing. Quantifi's release of this new model was covered in the September issue of Creditflux.

I agree that some transparency is needed here. Since many participants have shifted to stochastic recovery models, it has become very difficult to compare quoted correlations on a like-for-like basis. They are spot on with their view that the BNP Paribas model (Amraoui and Hitier) is the clearest implementation. The fact that it preserves the expected recovery rate consistently with single name bootstrapping - which is lost with a simple mark-down approach - was the main reason that we chose it for our implementation.

This is a welcome proposal, although there is no one model to cope with all structures, underlyings and attachment points. In our experiece, deals should be modelled using a number of different models and assumptions to get an envelope of possible values.