How do I model the randomness of recovery rate given default when pricing credit derivatives?
The standard reference is Anderson and Sidenius Extensions to the Gaussian Copula: Random recovery and random factor loadings. Random recovery proved necessary in 2007/2008 when you couldn't calibration standard one factor base correlation models. This paper discusses this, and might be an easier starting point than the Anderson and Sidenius paper.
CreditMetrics uses monte carlo simulation assuming a beta-distribution fitted to historical recovery rates.