The Case Against Standard & Poor's

Peter Eavis in a story that appears on page A3 of today's paper explains that the heart of DOJ's case against Standard & Poor's has to do with the computer models the company used to analyze both collateralized debt obligations and the underlying mortgages. The models were cooked, bogus, jiggered with to produce the high credit rating that the issuer wanted. As Eavis explains,

As home prices soared and buyers clamored for properties, banks began to churn out more loans and bundle them into mortgage securities. To analyze mortgages, S.& P. used a program called Levels version 5.6 at the time. S.& P. used that data to come up with its credit ratings for mortgage-backed bonds.

As early as 2004, S.& P. considered broadening the pool of loans in the model. The upgrade was intended to create a more realistic model called Levels 6.0, and S.& P. announced it was forthcoming. But the model was never released, the lawsuit claims.

Instead, S.& P. introduced a more modest upgrade in 2006, Levels 5.7, according to the suit. But the Justice Department contends that model did not provide an accurate picture of the loans. The suit said that an executive made a change to the model that would keep ratings artificially high. 

Ms. Mathis, the S.& P. spokeswoman, said the description of the adjustment to Levels 5.7 was inaccurate. “We are not aware of any changes made to the 5.7 model that were not analytically justified, nor that any changes were made by an individual as opposed to a committee,” she said.

As to the inference by the S&P spokeswoman that model adjustments were kosher as long as they were made by committee, Mary Williams Walsh and Ron Nixon address it in their frontpage story about Attorney General Eric Holder's announcement of the lawsuit to the media yesterday:

Remarks that S.& P. employees made in internal memos and electronic communications show that as early as spring 2004, certain executives wanted to change the firm’s rating methodology, but only after polling “an appropriate number of issuers and investment bankers” as to the “rating implications.” 

The idea of asking bankers what they thought about a change in the firm’s methods shocked some S.& P. analysts and executives, including one who fired back, “What does ‘rating implication’ have to do with the search for truth? Are you implying that we might actually reject or stifle ‘superior analytics’ for market considerations?” 

In May 2004, an analyst warned that S.&. P. had just lost to its competitor Moody’s Investors Service the chance to rate a very large deal by being too hard-nosed about the amount of collateral that would be required to get a good rating. More collateral would mean less profit for Mizuho, the bank putting that deal together. 

“We must address this now,” she said — otherwise the firm would lose more deals. 

The complaint describes a debate in 2004 and 2005 about whether S.& P. should change its model for rating C.D.O.’s and what effect the proposed changes might have on its business. The change was scheduled for July 2005, but before it could happen, an analyst sent an e-mail saying that according to the investment bank Bear Stearns, the older model “had been the ‘best’ ” at rating weaker pools of mortgages, compared with Moody’s and Fitch.

DOJ's case looks solid and easy to understand. In order to make garbage look like gold S&P ran the garbage through a computer model; it popped out as gold. If the computer model popped out something different, say, lead or bronze, a new computer model was devised. People can understand that. It's called fraud.