There are several dimensions to the problem of using the Gaussian copula. I agree with Clopinette that market prices are routinely used to provide a market outlook of the future. However, IMHO the main problem with this model is not that, but rather the fact that the incredible complexity of a CDO cannot be reduced to one correlation number. It is somewhat similar to attempts to express risk in a portfolio by a single VaR number. You cannot project a 125-dimensional space (actually it's more than 125 dimensions) onto a one-dimensional space without losing most of information. Who should be blamed for the disaster are not quants but traders that demand that models should be "intuitive" and "tractable".

QuoteOriginally posted by: ClopinetteI think it very naive to blame the mispricing of CDO on the gaussian copula:As bad as this model is (and it is), as all models, it only backs out market prices. In particular it uses what the "market" anticipates about all the names defaults.Well the fact it: the market was not anticipating much default about sub-prime names, otherwise the senior spreads would have been much higher.So don't blame it on the model: blame it on the blind market participant.But this is very typical: when something goes wrong, it has to be the quants' fault, never the fault of whoever has marked the curves....Models can be used for at least two things:1. Interpolating prices (from market prices of liquid products);2. Hedging / risk managment.Even if you can replicate exactly market prices with a model, it does not gives you necessarily a correct estimation of your risk and a good hedging strategy. So you can have a good model with respect to point 1 which is very bad with respect to point 2. Of course the "Gaussian copula model" or any other model is not bad in itself, but calling it a "CDO model" is very bad if it does not help you hedging your CDOs or estimating your risks.

Gaussian copula model is certainly bad for point 2, and apparently bad for point 1.

What is the 'right' hedge? When the synthetic CDO market was growing fast it almost felt that there was a 'silent agreement' with all dealers that the Gaussian Copula (plus some ad hoc mapping rules) was an accurate description on how tranches should interpolate and move. The answer to the important question 'why' is simply 'don't worry, that is just the way it is'.This was a quite fortunate setup for the dealer community as it allowed dealers to put most of their focus in printing trades with clients and happily showing risk management the risk was being hedged at the same time. With everyone using the same model to move tranche prices, risks were effectively being hedged, at least against relatively small movements in the underlying assets.If you wanted to be really cleaver and start trading with a 'better' model you would have been up for a big challenge. Your P&L volatility cold easily become enormous if the model resulted in significantly different hedge parameters. Even if you were 'right' and everyone else was 'wrong' you still could lose a lot of money by being 'right'. Patience would be needed and total devotion, plus hopefully a big move in the underlying credit spreads to prove your point.When the credit market started blowing up we saw a big shift in tranche valuations based on fear driven supply and demand. At this point both clients and dealers were more concerned with the question: 'Is this the right position to have on' instead of 'am I hedging my sensitivities'. Maybe a 'better' model would have made the progression to those new level much smoother with no specific need for big 'unexplained' jumps. In hindsight it's easy to say that the deltas were wrong but the underlying CDSs never seemed to move far enough for anybody to actually care.

QuoteOriginally posted by: ZiggyWhen the synthetic CDO market was growing fast it almost felt that there was a 'silent agreement' with all dealers that the Gaussian Copula (plus some ad hoc mapping rules) was an accurate description on how tranches should interpolate and move. The answer to the important question 'why' is simply 'don't worry, that is just the way it is'. Sounds like you're politely describing a bunch of sheep following each other like sheep.

So much derision for the 'copula model' (whether it's this copula or that), which I think is missing the point.Would you regard the correlation coefficient as a 'model'?

One of things that is still killing Wall Street might be this sort of analysis:http://www.rgemonitor.com/economonitor- ... orry_about So much for mister Felix Salmon.....