by Tanta on 11/29/2007 09:53:00 AM

Regular readers of this site will remember more or less constant outbursts of complaining about the lack of full-file due diligence in the securitization process. People look at data tapes and write contract warranties; people don't actually open up loan files and assess the accuracy of those data tapes, let alone go beyond the tape elements (quantifiable information like LTV and DTI and FICO) to qualitative aspects and conformity with "soft guidelines" (the rules of processing or evaluating documents or information that generates the quantities).

Well, glory be. Some analysts at Fitch rolled up their sleeves and sat down with a pile of loan files (45 of them, to be precise). What they discovered are the parameters of the the most innovative product the industry has ever offered, the NUNA (No Underwriting, No Accountability).

While we realize this was a very limited sample, Fitch believes that the findings are indicative of the types and magnitude of issues, such as poor underwriting and fraud, which are prevalent in the high delinquencies of recent subprime vintages. In addition, although the sample was adversely selected based on payment patterns and high risk factors, the files indicated that fraud was not only present, but, in most cases, could have been identified with adequate underwriting, quality control and fraud prevention tools prior to the loan funding.

I suggest reading the whole thing, if you've got a few minutes to blow on mortgage credit risk assessment arcana. What is happening here, although Fitch doesn't state it in these terms, is that someone went through a pile of loan files that were originally processed and underwritten in the "innovative" (cheap) way, and subjected them to processing and underwriting in the traditional (less cheap) way. As far as I can tell, the results here did not come from "extraordinary" levels of investigation or even much in the way of hindsight data (that is, by working back from events that took place after the loan was closed).

What this little exercise of Fitch's shows is that doing things like working through a credit report, looking carefully at the ownership of accounts (sole, joint, authorized user) and the relationship of tradelines to the information listed on the application (matching mortgage tradelines to real estate owned and matching general debt usage and spending patterns to income claimed) provides you with vital information that leads you to reject a mere FICO score as either a sole determinant of credit quality or as a "magic offset" that can let you ignore other weaknesses in the deal. No, really. All that is what we used to do before we started relying on FICOs and AUS.

But don't think this is just an anti-technology rant: 22% of the files Fitch looked at had a "HAWK Alert" right there on the original credit report, visible to anyone who reads English. What's that? It's a warning message that the credit repositories print on a report when some combination of facts or transactions trigger one of its potential fraud algorithms. It does not prove fraud, but it is designed to make your average human underwriter sit up straight and start poring over documents at a much greater level of detail and skepticism than might be usual.

Computers are great for this kind of thing: they help you put your work into "buckets" right off the bat. Any loan with a HAWK Alert should be immediately routed out of the "automation" pipeline and onto the desk of a senior underwriter. The technology helps you direct your human expertise where it really pays off.

Yeah, right, except that as Fitch notes, there is no indication in any of these files that anyone even noticed the HAWK Alert. And the problems that caused that Alert were, apparently, quite visible in the original file documents, if you looked past the FICO score to the details of the tradelines or the relationship of the tradelines to the loan application. (Loan app says borrower is a first-time homebuyer, credit report says there are existing mortgage tradelines. Duh.)

And of course the ridiculousness of the whole stated income thing is here on display, but Fitch never quite gets to asking the question raised by it all: if stated income is such high risk that you need to develop all sorts of processes and practices for testing it, contextualizing it, and subjecting the rest of the file to a fine-toothed review to compensate for it, what, exactly, is the benefit, in cost and speed, of doing it in the first place? At some level Fitch's analysis reads like a medical school textbook on using expensive fifth-generation antibiotics to treat staph infections caused by failure of doctors and nurses to wash their hands between patients. It's nice to know that can be treated, but it would be even nicer--not to mention cheaper--if hand-washing became more popular in the first place.

The fact is that many mortgage shops did, actually, put all kinds of practices and processes and "risk management control points" in place over the last several years to compensate for the risk created by their reliance on AUS, stated income, FICO scores, AVMs, and so on. For a lot of these shops, all that compensating was at least as expensive as just doing it the old way from the beginning. Certainly it remains to be demonstrated what kind of originator the files Fitch found came from: newbies who never learned the old ways and hence never saw the risk? Thinly-capitalized budget operations who just couldn't afford to produce anything except NUNA loans? Cynical players who knew the stuff was junk but who also knew that neither the security issuers nor the rating agencies would notice, and who also knew that the game of representations and warranties, properly played, would insulate them from having to take it back?

The whole thing really begins to take on an amazing Rube Goldberg quality once you refuse to accept the beginning premise--that these "innovative" ways of underwriting loans are a given, it's the compensation mechanisms that are the question. You do not have to believe that traditional human underwriting is perfect to wonder whether the cure is worse than the disease when it comes to compensating for automation.

(Hat tip to some awesomely cool dood)