Constant proportion debt obligations are, in themselves, complex products, but their mathematical complexity somewhat pales in comparison to the rating agency models which were developed to judge their default risk.
A rating model, in simple terms, is a simulation: data observed from the market (based on “methodological assumptions”) is fed into the model at one end and out the other end, a random outcome is produced. The “outcome” reflects whether the bond will, or will not, pay all of its coupons and return its principal. In what’s known as a monte carlo simulation, that process is repeated again, and again, and again - usually a million times - to give a whole range of likely “outcomes”. A rating is then awarded based on the number of those million outcomes which result in the bond defaulting. In the case of triple A, the simulation must show that of the million different outcomes, only a minuscule proportion are defaults.
The complexity with CPDOs comes in several parts: with an ordinary structured product, a model is built which statistically assesses the quality of the underlying collateral. For example, with a subprime CDO, a rating agency would look at past default rates and conservatively apply them across the collateral pool to obtain a figure for potential losses. Using past data, it’s fairly simple to extrapolate how that collateral will behave. In simple terms, the ratings for structured products like CDOs are based on fundamentals.
CPDOs, however, don’t make money from fixed-income collateral pools but from trading on credit default swap markets. A CPDO is essentially a dynamic and levered market bet.
The trouble with putting a rating onto such a product is that you have to make assumptions about how the market will move. Bankers, hedge funds and investors may have been doing that for a long time, but rating agencies haven’t been. For CPDOs, the rating agencies needed to develop a model which would accurately simulate the way CDS indices would move over the next 10 years.
At Moody’s the CPDO model - as with most structured product models - came in two parts: the dll and the CDOROM. The dll was the “black box” proprietary part: the secret mathematical model developed to spit out the rating.
The “error” in Moody’s code, which a Financial Times investigation revealed on Wednesday, was in the dll.
When Moody’s discovered the error they corrected it and found that this meant that standard “ABN-like” first generation CPDOs would lose up to four notches of their ratings. CPDOs rated after the correction, however, still achieved triple A.
In part, it seems this was because Moody’s made two simultaneous changes to their rating methodologies. These reduced the impact of the coding issue, say documents seen by the FT.
The changes reflected different methodological assumptions about the market. Most notably, the first change put a “volatility cap” onto Moody’s predictions for how the CDS markets would behave. This had the effect of discounting any scenarios spat out by the model which predicting large movements in price: in effect, the model was adjusted so it couldn’t predict the credit crisis.
More detail on that, in CPDOs triple A failure.
