Wednesday, May 21, 2008

Moody's Glitch Transposes A2 to Aaa

Well, one response by Moody's could be, how many of you guys know what A2 means anyway? Is it better, or worse, than A1? Aaa? When John Moody left S&P to start his own firm, he had to create a new scale, so S&P kept the more intuitive AAA, AA+ scale while Moody came up with his wicked nomenclature.

Anyway, the stock fell 15% today because Moody's admitted that a AAA security should have been 'several notches' lower but for a problem in their computer code. 'Several' means 'more than 3, but not many'. Let's say 5. So five notches lower is A2. Nice, as Borat would say.

The problem was with Constant Proportion Debt Obligations, instruments so complex Fitch wouldn't even rate them. An analyst at CreditSights in April 2006 noted that "though we cannot pinpoint exactly where the flaw in the ratings methodology is, there are a number of things to give us unease". [I should note that these things are infinitely less complicated than global climate models].

As mentioned in my post on why bank examiners should have access to pnl by line of business, a simple sniff test should have exposed this instantly. Even in 2005, these things had AAA ratings but 200 basis point spreads to Treasuries, while regular AAA debt had only a 20 basis point spread. Great opportunity, you might ask? Well, no. Moody's rates things based on an expected loss methodology. It puts collateral through scenarios, and notes the various trigger points that cause structures to collapse, in calculating the expected loss of various senior notes from these pools of collateral. That is, the expected loss on a AAA rated note on a pool of B rated credit is highly nonlinear, accounting for the fact that AAA pieces often have zero losses until several barriers are breached.

But Moody's is consistent, and so, the expected loss on these AAA rated senior pieces should be the same as the expected losses on all AAA rated debt, if they ran their models correctly. Now, one can imagine the market, due to illiquidity, might gives these things a 20 or 40 basis point premium, but 200? That's the market saying 'your models are wrong', which was, in effect, the absence of S&P and Fitch also said.

Agency ratings are best thought of in natural log scale, so that the natural log of the average default rates is about 1.5 unit different from 'grades' of AAA to B (.01%, .04%, .13%, .36%, 1.1%, 4.9%). So being off by several notches is material, because it goes from a 'notch' to a 'grade', which is discernible intuitively.

The key is, how can you trust something you don't understand? One easy way, would be not to. But this would leave one pretty catatonic. We rely on things everyday we don't understand, like how an airplane flies, or how my prescription drugs work, or even how my brain tells me to do something. And so we rely a lot of brand names and the reputation for integrity and thoroughness they represent. Thus, a hit like this should be very costly to Moody's, but that suggests the solution is endogenous, in that it should be in Moody's self-interest to correct this error.

In the meantime, when evaluating risk look at returns too. If they seem out of proportion to other returns on seemingly similar risks, you're doing it wrong.

No comments: