Page 176 - untitled
P. 176
THE CDO MACHINE
ties] group,” Gary Witt, formerly one of Moody’s team managing directors for the
CDO unit, told the FCIC. This approach would lead to problems for Moody’s—and
for investors. Witt testified that the underlying collateral “just completely disinte-
grated below us and we didn’t react and we should have. . . . We had to be looking for
a problem. And we weren’t looking.”
To determine the likelihood that any given security in the CDO would default,
Moody’s plugged in assumptions based on those original ratings. This was no simple
task. Meanwhile, if the initial ratings turned out—owing to poor underwriting, fraud,
or any other cause—to poorly reflect the quality of the mortgages in the bonds, the
error was blindly compounded when mortgage-backed securities were packaged
into CDOs.
Even more difficult was the estimation of the default correlation between the se-
curities in the portfolio—always tricky, but particularly so in the case of CDOs con-
sisting of subprime and Alt-A mortgage-backed securities that had only a short
performance history. So the firm explicitly relied on the judgment of its analysts. “In
the absence of meaningful default data, it is impossible to develop empirical default
correlation measures based on actual observations of defaults,” Moody’s acknowl-
edged in one early explanation of its process.
In plainer English, Witt said, Moody’s didn’t have a good model on which to esti-
mate correlations between mortgage-backed securities—so they “made them up.” He
recalled, “They went to the analyst in each of the groups and they said, ‘Well, you
know, how related do you think these types of [mortgage-backed securities] are?’”
This problem would become more serious with the rise of CDOs in the middle of the
decade. Witt felt strongly that Moody’s needed to update its CDO rating model to ex-
plicitly address the increasing concentration of risky mortgage-related securities in
the collateral underlying CDOs. He undertook two initiatives to address this issue.
First, in mid-, he developed a new rating methodology that directly incorpo-
rated correlation into the model. However, the technique he devised was not applied
to CDO ratings for another year. Second, he proposed a research initiative in early
to “look through” a few CDO deals at the level of the underlying mortgage-
backed securities and to see if “the assumptions that we’re making for AAA CDOs are
consistent . . . with the correlation assumptions that we’re making for AAA [mort-
gage-backed securities].” Although Witt received approval from his superiors for this
investigation, contractual disagreements prevented him from buying the software he
needed to conduct the look-through analysis.
In June , Moody’s updated its approach for estimating default correlation, but
it based the new model on trends from the previous years, a period when housing
prices were rising and mortgage delinquencies were very low—and a period in which
nontraditional mortgage products had been a very small niche. Then, Moody’s mod-
ified this optimistic set of “empirical” assumptions with ad hoc adjustments based on
factors such as region, year of origination, and servicer. For example, if two mort-
gage-backed securities were issued in the same region—say, Southern California—
Moody’s boosted the correlation; if they shared a common mortgage servicer,
Moody’s boosted it further. But at the same time, it would make other technical
