Moody’s, Rating Models, and CDOs

On page 140, we write this:

“In 2004 and 2005, both Moody’s and Standard & Poor’s modified their rating models in ways that made it easier to give higher ratings to CDOs, helping extend the structured finance boom.”

The source for that sentence is a Bloomberg article entitled “‘Race to Bottom’ at Moody’s, S & P Secured Subprime’s Boom, Bust.” The article begins this way:

“In August 2004, Moody’s Corp. unveiled a new credit-rating model that Wall Street banks used to sow the seeds of their own demise. The formula allowed securities firms to sell more top-rated, subprime mortgage-backed bonds than ever before.”

Further down, the article focuses on a rating model introduced in August 2004 by Gary Witt, which shifted from the older “binomial expansion technique” (BET) for modeling diversity in a portfolio of assets to a “correlated binomial” technique, and quotes other sources saying that the effect of the new model was to boost the ratings of CDOs.

Gary Witt is now a professor of statistics and finance at Temple University, and he sent me the following information by email:

  • The new model would have increased the projected losses for AAA CDOs relative to the BET approach, which might have implied lower ratings, but not higher ones.
  • The model introduced in August 2004 was not actually adopted by Moody’s for rating CDOs based on RMBS (residential mortgage-backed securities) or ABS (asset-backed securities, a category that often included subprime mortgage-backed securities).
  • Instead, in 2005 Moody’s adopted the normal copula approach (favored by the investment banks).

So, if Witt is correct (and I have no reason to think he isn’t), the underlying article we used was wrong. There is still the question of what impact the 2005 change to the normal copula approach had. (Felix Salmon has previously criticized this type of model.)

Witt’s opinion is that the new model on balance did not make it easier to give higher ratings for CDOs. The BET had assumptions about independence that were clearly inaccurate by 2005, and the new model was an improvement. Still, Witt acknowledges that it’s not an open-and-shut case, in part because the models take different approaches to measuring correlation. For one thing, introducing a new model induces investment banks to behave strategically and game the new model, so it doesn’t make sense to simply take a given CDO and rate it using the two models; in practice, the banks will create different CDOs that are influenced by the properties of the two models.

So on balance, the sentence at the beginning of this post isn’t supported by the source we cited (at least when it comes to Moody’s).

%d bloggers like this: