Structural product rating and sensitivity to mismeasurement

The latest issue of the Journal of Economic Perspectives has a nice symposium on the early stages of the current credit crunch. While many arguments developed therein are nothing new, which in some way proves their validity, I found the piece by Joshua Coval, Jakub Jurek and Erik Stafford to be particularly interesting.

They look at how the credit rating agencies tried to evaluate structured products. They are different from the typical bonds they were used to rate in that they are pools of underlying assets (for example securitized mortgages) that are then split by level of debt seniority. This allows the most senior debt to, in fact, become more secure than the portfolio. But it still has a risk of default, and this risk depends on the correlation of default of the underlying assets. This is where it becomes tricky.

First, it appears ratings are extremely sensitive to this correlation. The yearly default rate for the ten classes between AAA and BBB- range from 0.02% to 0.75%, truly minuscule and empirically very difficult to distinguish. And the measurement problem, already present with regular bonds, is amplified by the correlation uncertainty, a measure credit agencies were not used to. Second, it appears these correlation have been widely underestimated. As reported at other places, the formula was essentially wrong. Rating agencies were used to estimate simple distributions (as for bonds), not joint distributions. And the poor geographic diversification of these products was not recognized. Thirdly, structured products have been further repackaged, amplifying further any of the problems mentioned above.

So all-in-all, it seems the rating agencies dropped the ball in a major way. And this is before one realizes the other problems related to the structure of that industry.