by Tanta on 8/11/2007 08:51:00 AM

I just got around yesterday afternoon to reading the full announcement that went with S&P's recent negative rating actions on Alt-A deals. I know it's not as exciting as Fed repos, but periodically we are allowed a little "No, really?" at the expense of a rating agency:

In late 2005 and 2006, mortgage origination underwriting guidelines expanded rapidly, which allowed the proliferation of layered risks within the Alt-A market. This combination of multiple risk factors for a single loan is the principal driving force behind the deteriorating performance of the 2006 vintage. Historically, the presence of high FICO scores within a loan has proved an effective mitigant to increased risk factors elsewhere, such as higher CLTVs. However, the increase in recent delinquencies across all FICO bands indicates that a borrower's previous credit performance is less
predictive of stronger performance for loans with increased risk layering. This emerging delinquency performance has prompted us to reduce our emphasis on FICO score as an offset to layered risk.

Insofar as FICOs are accurate measures of past performance, high scores indicate borrowers who have managed credit wisely in the past. Put those borrowers in unwise credit terms, and they perform just like people who have managed credit unwisely in the past. Glad we got some real-time empirical data to prove that. Sorry about your global financial crisis.

There's more:

Recent delinquency data also indicates a need to adjust default expectations for certain purchase loans. These loans are underperforming our initial assumptions, particularly when combined with high CLTVs. The performance related to purchase loans is unprecedented in historical data. We will increase our default expectations for the increased risk at high CLTVs, particularly those with CLTVs that exceed 90%.

It is, of course, perfectly true that all "historical data" I am aware of has shown lower risk for purchase transactions than for refinances; somewhat lower than no-cash-out and significantly lower than cash-outs. Of course there has always been a bit of a problem around the "well, controlling for CLTV, that is" part. As with the FICO thing, we only just got ourselves a big database of loans with nutsy CLTVs for all loan purpose types.

The thing is, S&P isn't the only one with a model that has been giving extra credit in the risk-weighting to "purchase" transaction types; just about everyone has. Credit models, pricing models, due diligence selection protocols--they've all included the "purchase benefit." The point is to ask why it is no longer a "benefit," and the CLTV issue is only a part of that. Or, at least, there's more to the CLTV issue than its value relative to historical lending patterns.

The fact is, historically appraisals for purchase-money transactions were the most reliable. Time and again you could test them and see this. We have always believed that it had something to do with the fact that in a purchase transaction, you have a sales price to work with. There was a vague sense among us that with a buyer and a seller out there behaving like Econ 101 says they behave, the sales price--and the comparable sales prices--would ground a purchase appraisal in some kind of "reality." It's not that all refi appraisals are bad, but that they are, as I've said before, inevitably a kind of "mark to model." Purchase appraisals are supposed to be "mark to market."

Funny how some people's models have worked out better than a lot of people's markets, isn't it? Value, of course, is not simply equal to price, and prudent lenders wouldn't be messing around with the time and expense of appraisals if it were. Everything you read about appraisers being pressured to "hit the number" boils down to an industry trying to force "value" to equal "whatever dumb offer someone can be crazy enough to make today." S&P wants to see this as a CLTV issue, and surely it is wise to stop making sunny assumptions about 100% financing, but the fact is that 100% financing works as long as the numbers keep going up.

We were talking yesterday about jumbos and conforming loans and their relative risks. Traditionally, lenders always required two separate independent appraisals for higher-end properties. For years, the cutoff was $650,000 or thereabouts; it sneaked up to $1,000,000 during the boom. I can remember reading a major Alt-A conduit's guidelines in 2004 or so and discovering that their appraisal standards depended--get this--on LTV: for loans over $650,000, a second appraisal would be required if the LTV were over 70%. It said that in the published guidelines. It made sense to a bunch of credit analysts that you could use "V" to decide how you were going to determine "V." Certainly, using a dollar-amount rule can sometimes seem arbitrary. But count me in the arbitrary group.

We are learning here that what are called the "soft guidelines"--all the rules and procedures of a lender that are not easily quantifiable in numbers that can be plugged into a computer model--are making a difference. OK, well, some of us have been insisting for years that this is the case, but the world that wants cheap, fast credit analysis of huge pools of loans without loan-level due diligence or highly-complex analytical models (say, ones that have more than Fitch's famous three doc types), is apparently in wounded-innocence stage. No wonder we'd rather be stunned and surprised by a weekend's worth of Fed repos.