CLO default correlation

blaugranabhoy · June 20th, 2011, 9:14 am

Hi,I'm trying to model a CLO structure as simply as possible and have taken the approach of modelling default probabilities as Bernoulli random variables, and then applying a correlation effect between these entities. On a very high level, I was wondering how this correlation is achieved. I haven't simulated binary variables previously, having only looked at equity/index simulation with correlation.Any help would be very welcome.

Costeanu · June 23rd, 2011, 1:10 am

You should read about the "gaussian copula" credit model. It's quite famous, and it landed a place of honor on Paul's "Name and shame" list (some people think the gaussian copula caused the crisis). Observing a Bernoulli variable X (1 with p, 0 with 1-p) is the same as observing a normal variable Y and comparing it with a treshold N^{-1}(p). You could call the variable Y the gaussian driver of X. Now correlating a number of Bernoulli variables can be done by correlating their gaussian drivers. Best,V.

blaugranabhoy · July 9th, 2011, 9:09 pm

Thanks Costeanu...hadn't checked boards for a while (so didn't see your response) but had used the approach you suggested for my model. Appreciate your response.