I’m not saying it’s not highly liquid now, or that you can’t get plenty of data now. The data question came up regarding the integrity of information prior to 2001.

8:52 AM said the dealers would have the information, which appears to be confirmation that there isn’t much older third party data. Having worked more than once with data in OTC markets, I can tell you the dealers’ own trade information (when the market isn’t deep enough to have good third party reporting) often isn’t very satisfactory.

I believe the CDOs mentioned are standardised tranches of the CDX and ITRAXX indices.
There are numerous platforms out there on which to trade single name and indices – CreditEx, Brokertec etc – I assume all side side have this (ours do) – it is a very liquid market

http://www.creditfixings.com/information/affiliations/fixings/itraxx_fixings.html

The article did not indicate that there has been price pressure in CDOs of corporate debt. Indeed, until the comments on this piece, I’ve never seen any mention of CDOs composed solely of corporate debt.

There hasn’t been all that much corporate bond issuance, and there is good demand for investment grade names. Why would you need to bundle it into a CDO? I can see leveraged loans, but I was under the impression they mainly went into CLOs.

Nevertheless, the media has made clear that CDOs are very heterogeneous, so a CDO could clearly contain some corporate debt exposure.

While the CDS market started in the mid 1990s, volumes didn’t become significant till around 2003. And this is an OTC market. The only data a dealer would have in the absence of being able to buy data from a service is its own trades. Hardly representative in a not-deeply-traded market, particularly if the traders’ pricing was influenced by a need to change the shape of their book.

CDO of ABS => hedge with ABX, CDO of corporate debt => hedge with CDX/iTraxx.

Regarding the paucity of data for CDS indices, single name CDS have been traded since, what, the mid-90s? Presumably some dealers have much more historical data than, say, Markit does.

wrt to credit risk being too high, there was a similar story with BAA recently

http://www.noelwatson.com/blog/PermaLink,guid,f5c355be-b7ad-4e93-96b1-4dee8b737444.aspx

I believe the CDOs they are talking about here are the CDX index tranches. In Europe the equivalent is the ITRAXX Europe. People tend to buy the index rather than the underlying constituents when buying credit protection as there is more liquidity (CPDO unwinds are rumoured to be doing this). This drives the index wider and the underlying single names drift wider as the arbitrage disappears. Looking at historical default rates the indices appear oversold – but there could be trouble ahead.
The article mentions correltion over 100%. I assume a Gaussian model is being used – this doesn’t accurately reflect the credit markets (fat tails). There may be better models – Levy for example

http://www.noelwatson.com/blog/PermaLink,guid,136d4b18-8d46-4687-8a83-0ef2f97ab805.aspx

Thanks. One point I should have clarified is that the increased hedging (according to the article) was driven by worries about CDO price declines. That presumably would have been most acute in CDOs with some or a lot of real estate exposure.

A general practice was that you mapped your CDO portfolio onto ITRAX/CDX in some way (say if you had only 5 different entities, you could say that the correlation was almost like ITRAX with some small mods…)

The problem is more with model in general, and that the correlation is a very opaque and vague thing, even more so than volatility.
You can track vol of an option on company’s stock, maybe even for years. You can’t track their default correlation by definition – because if you’re interested in it, they haven’t defaulted yet. In a way credit market correlation has the ultimate survivorship bias.

Also, the correlation in the models is simplified so that instead of tracking say a matrix of 125×125 correlations you track only a single “super correlation” (base correlation). Of course, by doing so you loose information, and as a result can get very strange outcomes.
When Ford/GM were downgraded few years ago, it was the first time the correlation models broke (the implied correlation from the model was nonsensical) and people had to fix them in a hurry. Of course, this debacle was later presented as the victory for the model, with people saying it has survived its big test. Yeah, right, said I.

Of course, even if the model worked as proposed, hardly anyone was paying attention the the risk profile of the model, which was off-the-cliff. That is, you have virtually no risk for a long period of time/stress, but when it goes to the extreme right, the value just falls of the cliff. Which is exactly what happened.