I was curious about the accuracy of our credit reserve model as a function of 
the
number of simulations we use. This question, I think, is not so important 
when we 
calculate credit reserve, because the assumptions underlying our model are 
pretty 
rough anyway (here I mean the assumptions regarding price processes, 
correlations, etc.)

This question becomes more essential when we talk about calculating 
sensitivities of the 
credit reserve to various factors. When the magnitude of the sensitivity is 
comparable
to the accuracy of calculation of the credit reserve, what is the accuracy of 
such sensitivity?
  
I performed a numerical experiment where I calculated the expected loss
for a simple portfolio with one counterparty (SITHE IND POWER) for different
number of simulations (10, 100, 1000, 10000, 100000) using old research 
credit model.
You can see how the result converges and the relative error (compared to the 
result for 100000 simulations which is assumed to be the most accurate) in 
the 
attached file.

Tanya.