Featured Article

Investor Beware Of Subprime Risk Models Dow Jones Newswire, July 18, 2007

NEW YORK (Dow Jones)--Investors surprised by the subprime mortgage meltdown
are getting a jolting reminder that predicting the future - even with high-tech
tools - is an inexact science.

Specifically, it seems that investors now evincing shock at higher defaults
and delinquencies among subprime borrowers and the losses they can inflict on
their mortgage bond holdings were relying too heavily on credit models - with
all their flaws and limitations - to assess risk.

If the models weren't flashing red, there were still plenty of other warnings
such as published research, news articles and discussions at industry
conferences over the past two years about what would happen to subprime mortgage
credit should home prices start to fall.

"The mistake that I think people make over and over again, and it pains me, is
to ascribe unreasonably high degrees of certainty to the answers they get out of
models," said Mark Adelson, head of fixed-income research for Nomura Securities
International in New York.

Most investors use some kind of mathematical model to assess the risk of
default in complicated subprime mortgage deals, which generally are sliced into
various tranches of risk in which AAA-rated slices are cushioned from potential
losses that are first absorbed by the lower-rated ones.

Most investors also are aware that credit-risk models, including those
developed by rating agencies, private companies and investors, have limitations
and are constantly being updated, Adelson said.

Models are a way of describing a series of calculations based on inputs about
loans, such as the size of the loan, the borrower's credit score and the total
amount of debt versus the total amount of income a borrower has. Once all the
data has been synthesized, the model will give values that will help investors
predict things like the probability a given loan or pool of loans will default
and what the size of the losses is likely to be.

Fatal Flaws

One limitation: Credit models tend to rely on historical data, and if the
historical data is thin or nonexistent, accuracy can be lower.

Loose lending standards in recent years with loan programs that allowed
borrowers to put no money down and also to state their income and assets with no
documentation were "really something the market had not seen before," said Tom
Warrack, a managing director in the residential mortgage-backed securities group
at Standard and Poor's.

Therefore "the historical data on performance that we had to review was
somewhat lacking in some of these new loan programs," he said.

However, especially in these circumstances, Adelson said: "Common sense tells
you you are not going to achieve a really high degree of certainty" from the
answers given by models.

Standard & Poor's and Moody's Investors Service last week downgraded ratings
for several hundred subprime-backed mortgage bond transactions on watch for
downgrades.

S&P has since announced changes to its model for rating subprime bonds,
including making higher assumptions about default expectations for certain types
of common subprime loans. Moody's also announced changes to its model on a
conference call last week.

Other types of flaws in models can be more technical. For example, for loans
with several different layers of risk - zero down payment as well as no
documentation of a borrower's income or assets, for example - some models may
have been adding these risks rather than multiplying them, one fixed-income
portfolio manager said.

But having all these different levels of risk together increases the overall
risk of a loan in a way that is greater than just looking at all the risk
factors in isolation and then adding them.

What that means is that if negative effect A would lead to $2 million in loss
expectations and negative effect B would lead to a $3 million loss, a model that
added the effects would project a $5 million loss, and a model that multiplied
them would project a loss of $6 million.

Also, models could simply get certain assumptions about the market wrong, the
portfolio manager said. Among subprime borrowers, models may have underestimated
the number of first-time buyers purchasing homes, he said - and first-time
buyers are known to be riskier credits than those who have previously owned
homes.

A Broader Lesson To Be Learned


But whatever flaws may be in the models, a broader lesson to be learned from
the current subprime troubles is that relying too heavily on this type of
analysis to make investment decisions is a bad idea.

While there were "certainly things wrong with the models," said Andrew
Davidson, president of Andrew Davidson & Co., a firm that develops models for
analyzing mortgages and mortgage-backed securities, it is unfair to lay the bulk
of the blame on rating agencies and their methods.

"Investors are supposed to perform their own due diligence," Davidson said.
"It's your responsibility as a portfolio manager to evaluate the instruments you
invest in."

With home prices falling in many metro areas across the U.S., Adelson said:
"If you expected BBBs would not be downgraded...I'd like to ask what you were
smoking. This ain't news. If they weren't being downgraded now, wouldn't that
mean they should all be AAA?"

The fact of the matter is that the premiums investors could earn for investing
in subprime mortgage-backed bonds got relatively low in recent years, even for
some of the lower-rated bonds, he said.

And "the essence of fixed income investing is earning a premium" above, say,
Treasury yields, said Adelson. "If that premium is very small, you'd better be
able to gauge risk accurately."


-By Danielle Reed, Dow Jones Newswires