In Depth Analysis: CalculatedRisk Newsletter on Real Estate (Ad Free) Read it here.

Sunday, June 24, 2007

Murk to Muddle

by Tanta on 6/24/2007 08:17:00 AM

naked capitalism has a nice post up this morning that combines all of my favorite themes: models, valuations, risk management, and what the hell is the matter with the New York Times today?

I am not a Times Select subscriber--I get most of my useful business analysis from Opera Weekly--so I don't spend enough time picking on Gretchen Morgenson since she went behind the pay wall.

But Yves is out there doing the heavy lifting:

Back to Morgenson. She got a very important issue wrong:


Officials at ratings agencies have said in the past that their ratings reflect their estimates of future performance, not market pricing. So the agencies are also marking to model.

Ratings agencies are not Bloomberg terminals. They provide ratings.

There aren't many people who are more cynical than I am about the rating agencies, but a whole lot of people need to memorize those last two simple sentences. We can argue all day long about whether these bond ratings are sufficiently stress-tested, whether downgrades are too slow, whether the rating agencies conflate "collateral problems" and "structural problems" in their analysis. The whole sorry issue of their fee structure and non-arm's-lenth relationships with the bond underwriters could keep us talking for weeks. But if you want to tell me there's an "observable market rating" to which the agencies should be "marking," you'll have to tell me who writes your prescriptions.

To suggest that the rating agencies are "marking to model" is mind-numbingly dimwitted. This is just like The Great FICO Uproar: everybody wants to get all fired up about whether FICO scores are "inflated" or "manipulated" or what have you, as if FICOs are somehow supposed to be something other than just scores produced by a model.

News flash: There are good models, bad models, and ugly models. There are transparent and opaque models. There are stress-tested and untested models. They're all models. And if they're claiming to model probability of principal loss via default of the underlying collateral, that number you get at the end isn't a dollar price. The number you get at the end could be an input into a pricing model, to be sure. But would you really want to claim that an apparent failure of your pricing model is caused solely by one credit model-generated input failing to correlate to a future market price? If so, you aren't using a "pricing model." I don't doubt that there are some stupid investors out there who have been acting as if a certain credit rating--on a mortgage loan or a CDO tranche--guaranteed a certain market price. But the technical financial-accounting term for those people is "doofuses."

Maybe we could just have a moratorium on anyone using the phrase "mark to _____" for rhetorical flourish until we all get a little clearer on the concept.