Friday, February 5, 2010

CPDOs, a structured finance post-mortem

Original posted on FT Alphaville by Tracy Alloway:

Does anyone remember Constant Proportion Debt Obligations, or CPDOs, for short?

The erstwhile structured finance-darlings made headlines in 2008, as they hit their cash-out triggers en mass, and despite their triple-A and AA-ratings. They also featured prominently on FT Alphaville.

We bring them up again only because the Federal Reserve has put out a very interesting working paper on the subject of CPDOs and ratings methodologies. The whole thing is worth a read for some good old-fashioned acronymic geekery, but we’ve summarised the basics below.

Remember that these things, which basically used credit-default swaps (CDS) to bet on company creditworthiness, began popping up in the summer of 2006. ABN Amro’s Surf was the first to be issued, with an AAA rating. It went belly-up in October 2008, after corporate CDS spreads widened and volatility zigzagged in the midst of the credit crisis.

Surf is just one example of a CPDO (a UBS deal was actually the first to default — just eight months after Moody’s rated it triple-A) but it’s indicative of the market as a whole. The fact that it was named “Deal of the Year” in 2006 adds poignancy to the idea that these things were basically the poster children for what was wrong with the structured finance market in the years before the financial crisis.

Anyway – it’s the ratings methodology and modeling that the Fed paper is concerned with here.

The basic idea being that none of the CPDOs, or agencies, had anticipated the below:

The great bulk of CPDO issuance took place in the fall of 2006 and spring of 2007. In this period, corporate credit spreads were at a cyclical low point and volatility in credit spreads had been subdued for nearly three years . . .The first tremors of the credit crisis reached the corporate CDS market in the summer of 2007. The CDX index spiked to 81bp in early August, fell to 45bp in early October, reached 85bp in late November and thereafter increased steadily through mid March 2008, when it reached 193bp. Spreads then fell sharply to 86bp in early May, but soon were increasing again. The CDX broke 279bp on November 20, and finished the year at 195bp. The iTraxx spread followed a very similar pattern . . .

You can see the rather dramatic movements in the credit derivatives market in the chart below:


Ratings agencies who were rating CPDOs (the paper focuses on Moody’s and Standard & Poor’s) meanwhile, had not been taking into account the possibility that spreads would reach such levels. In fact, they were modeling such a development as having a probability of virtually zilch.

Here’s what the paper says:

For both S&P and Moody’s, the rating of CPDOs was driven by models that assigned effectively zero probability (of order 10^−7 or less) to reaching the levels of spreads actually realized in the CDS markets in late 2008. This spike in spreads would, to the best of our knowledge, have triggered default and large losses for every CPDO note that had managed to survive to that point. All the issues in question had been rated AAA at origination in late 2006 or early 2007.

So what, you might say? No one could have predicted the credit crisis, and it would have been unreasonable to ask the ratings agencies to take into account such `fat tail‘ spreads.

That’s true to an extent. But here’s the killer bit from the paper:

. . . we find that the agency models failed under more mundane conditions. The models assigned very small probabilities (well under one basis point) to reaching the spread levels seen in the second half of 2007. Credit market conditions at the time were qualitatively comparable to those of the previous downturn (2001–03), and therefore should not have been relegated to the extreme tail of possible outcomes. Even during the quiet pre-crisis period, the models significantly understated kurtosis. Thus, even if the crisis is viewed as an unforeseeable regime shift, the model underpredicted the likelihood of realizing high spreads within the pre-crisis regime.

Former FT Alphaviller Sam Jones was instrumental in the ratings agency-CPDO discussion, uncovering a software error in Moody’s CPDO-ratings sytem in May 2008. In fact, the paper makes more than a little mention of the scoop.

Notably though, it does not attribute the whole of Moody’s CPDO-ratings to the computer error:

In May 2008, the Financial Times reported that Moody’s CPDO ratings may have been affected by software coding errors. If it were the case that the poor performance of the agency ratings could be attributed to coding errors or subsequent model modifications, then our analyses would be of little interest . . . our conclusions appear to be robust. Furthermore, there have been no reports that coding errors played a significant role in S&P rating of CPDOs.

If FT Alphaville readers ask nicely, Mr Jones might revisit the CPDO-discussion later on.

No comments: