Editor's note: DataQuick®, a provider of advanced real estate information solutions recently announced that Gordon Crawford, Ph.D. had joined their firm as vice president of Analytics. MND doesn't normally note such appointments, but it caught our eye that Crawford who until recently was vice president of Mortgage Research at Fannie Mae, is an expert on credit scoring performance models and spoke about that topic at a recent Mortgage Bankers Association Risk Conference. He was kind enough speak with us on the topic while preparing for that meeting.
While a lot of us can now actually sing the words, the original screwball television commercials about credit scores marked the first time many people ever heard the words. Dr. Crawford told us that until recently credit scoring was sort of a "black box" for consumers. Compared to credit reports, scoring is a relatively young credit tool, and even after consumers learned they had a score, many had no real idea of what that meant.
In the last few years FICO, the main purveyor of credit scoring technology, and other providers have worked to educate borrowers, providing information about what goes into credit scores and consumer guides on how to improve them. Another recent consumer innovation is the ability to do credit score simulations online.
Using one of these simulators a consumer can, for example, test the effect of paying off one debt entirely in contrast to paying a portion of another or whether opening a new line of credit will have a positive or negative effect. This, Crawford said, is a double edged sword. Credit scoring companies now know that these simulators have given consumers the ability to game the system and must take this into account whenever they make changes to their models. "People should know what is in their score and how to improve it," he said, "but being able to manipulate it can invalidate the scorer's model."
In two respects credit scoring is not a static science. From the consumer standpoint, Crawford said that what were considered acceptable credit scores before then started to decline in 2002. A score of 700 used to be considered good, then 660; then suddenly it was 580. That trend flipped in 2008 and since then scores have ramped up to what is now a 760 to 780 average for what is considered a prime score. "We now have a strange situation," he said, "with very low interest rates but very tight access to credit. What used to be considered a good score, 720, is now one where lenders have to propose alternative programs. We have swung from a standard that was way much low to one that is probably too high."
From a lender standpoint credit scoring doesn't stand still either. Crawford said that vendors frequently update scoring models and the industry is challenged by this because it is not easy to adapt. All credit scoring is predicated on a FICO model and changing one necessitates changing or at least tweaking others. Existing pricing and underwriting models are also already calibrated to old scoring, making it costly to put new ones in place. Consequently lenders want to see substantive changes or ones that provide measurable benefits before they undertake the trouble and expense of training their people and changing their systems in order to accommodate them.
A case in point is a new model recently introduced by FICO which introduces data from a new CoreLogic credit report into the scoring and thus gives weight to such information, according to FICO "as property transaction data, landlord/tenant data, borrower-specific public data, and other alternative credit data. FICO maintains that the new model raises the scores of many borrowers - about 3 percent more individuals would score above the current 715 median - and has a predictive value of risk performance 7-1/2 times greater than current models.
Crawford said "Newer measures to beef up thinner files need to be proven out. If one lender adopts a new model and others don't follow then it is hard for him to transfer a loan or its servicing, sell the loan on the secondary market, or communicate about that loans and its risk. He risks losing common measures with other lenders."
Besides, Crawford said, lenders are happy right now with the tighter underwriting standards. Behind their concerns about any loosening of them is an uncertainty both about the capital levels that will be required for holding mortgages and about the future of Fannie Mae and Freddie Mac. "Until they have a higher comfort level they won't be willing to accept more exposure to risk."
He said that lenders, of course, use credit scoring as one in a whole arsenal of tools when they evaluate risk; appraisals, owner occupancy status, and other factors all enter into the equation. In addition, Crawford said, credit scores are intended to predict success with a loan, not necessarily to predict success with a mortgage loan. For this reason many lenders beef up scores by taking specific fields from them - for example the utilization rate of revolving debt - and wrapping them into their underwriting, thus multiplying the weight they carry.
Crawford's role at DataQuick will involve developing all models and analytics, including home price indices, automated property valuation models and loan performance models. While at Fannie Mae he worked on performance models used to determine loss allowance measures and pricing. Crawford earned a Bachelor of Arts in Economics from Brigham Young University and both a Master of Arts and a doctorate in Economics from the University of Rochester.
