When p>0.5, use the survival version of the tree instead of the default version, and use 1-p.

There aren't any more detailed papers. Lehman has a similar one.

SC is quite correct. You need to take the dealer's spreads and your model and back-out your own base correlation values. They won't be particularly close to the dealer's base correlations.Base correlation is primarily an interpolation tool to price non-standard tranches.

The "right" price is the market price. Your model should fit it by implying correlation from price (really, base correlation).

I know of nothing that exactly matches, but here are some references that might help.John Hull and Alan White "Forwards and European Options on CDO Tranches"Chapter 18 in A. Bomfim "Understanding Credit Derivatives..."Chapter 10 in J. Felsenheimer "Active Credit Portolio Management"

The relationship you describe should be the case. It is definitely not an issue with the copula model, so there must be some flaw in implementation.

If you do a Google search, search on SSRN, or look at DefaultRisk.com there are quite a few papers that discuss various methods of pricing CDOs

cumsum(a) and cumprod(a)

If you have the Statistics toolbox in Matlab you can use mvncdf. For VBA you can get code from http://www.vbnumericalmethods.com/math/. Or you can read Hull and code it yourself

About 2 years ago I successfully compiled Quantlib and Boost on Cygwin with gcc. There were some issues but they were resolved and messages were posted on the Quantlib list.

DD,Are you maintaining an errata for this text?

Send me a personal message if you want some papers from the agencies on their methodologies.

No. You must have some way for Excel to communicate with Matlab, and those are the only basic 2 alternatives. At least with the compiler you can make an add-in.

Well, for starters, you do not have enough input arguments. Try adding 2 more [] before 'confun' corresponding to LB & UB.

<t>It is typical to use 48 nodes for the 1-factor Gaussian copula. The examples in the paper are simplistic and I would not recommend choosing the number of nodes in general based upon it.I wouldn't worry about matching results from Hull exactly. There are many little details that are not explained ...