Within hours of each other Laurie Goodman, co-director of the Urban Institute's Housing Finance Policy Center, and Melvin Watt, Director of the Federal Housing Finance Agency, expressed quite different views on the need for an immediate change in the credit scoring models used in most mortgage underwriting.

Goodman, writing in the Policy Center's blog, talks once again about the tight credit box for mortgage financing.  The mortgage market, she says, is taking less than half the risk it took in 2001, a period of reasonable credit standards, and less than a third of what it took in 2006, when standards were what proved to be disastrously loose.  The tightness is greatest regarding credit scores. UI attempted to quantify this a while back and updates that exercise it in this article.  Had the credit score distribution remained the same as in 2001, they say, there would have been 1.1 million additional loans made in 2015, and a cumulative 6.3 million additional loans from 2009 to then.

Goodman says the credit score model used by Fannie Mae and Freddie Mac (the GSEs) is outdated. Both essentially require lenders to use the FICO 4 scoring model, based on models estimated in the late 1990s. For his part, Watt cited, in his speech on Monday, GSE's refinements to FICO 4 as one argument against an immediate change.  "Credit scores are only one factor the Enterprises (GSEs) use to evaluate loan applications and the Enterprises currently use the same or even greater levels of credit data in their underwriting systems as the credit scoring companies use."  

Newer models, Goodman references the FICO 9, and VantageScore 3 (VantageScore 4 will be released soon), have several advantages over the data used in the 1990 version which, first, was much less granular. Older credit records treated student loan debt as installment debt and made no differentiation between first and second mortgages.  There is also much better information in the newer models on student loan debt performance and how it effects that of other debt. She notes that in the late 1990s outstanding student loan debt was under $100 billion, today it is $1.34 trillion. Federal Reserve data shows that, from 2003 to the first quarter of this year, total consumer debt rose by 76 percent while student loan debt grew by 458 percent.  

The older FICO 4 family of models included paid collections while more recent FICO and VantageScore models ignore them as having limited probative value. Newer models also differentiate between unpaid medical and unpaid non-medical debt, weighting medical debt less heavily as it has been found less predictive of future performance.  This difference can affect a lot of borrowers as UI estimates that 35 percent of consumers with a credit score have a debt in collection.

Goodman points out that FICO 4 has a model for each of the three credit bureaus, each estimated in slightly different time periods between 1995-2000.  Lenders are instructed to obtain a credit score for each loan from each of the three credit bureaus, using the middle score for lending/pricing purposes. More recent FICO and VantageScore models use identical time periods and more closely or completely align the models. The GSEs use the FICO 4 model for both determining the loans eligibility for purchase and loan level pricing adjustments. Fannie Mae's Automated Underwriting System does not use a credit score but pulls credit data directly while Freddie Mac's supplements the credit score with additional credit data.

In addition, FICO 4 credit scores influence lender decisions about which loans to originate for potential sale to the GSEs. Goodman claims there is widespread acknowledgement that this reliance on an outdated FICO models should change.

She acknowledges in detail the work Watt said FHFA has done in studying the costs and benefits of a new scoring model and the efforts of the GSEs to improve access to credit and adds that, in their 2017 Scorecard, the agency directed the GSEs to "Conclude assessment of updated credit score models for underwriting, pricing, and investor disclosures, and, as appropriate, plan for implementation."  In addition, she says, HR 898, "The Credit Score Competition Act of 2017," and an expected companion bill in the Senate, would encourage the GSEs to consider alternative credit risk scoring models when making mortgage purchasing decisions, and to establish and make public their procedures for validating and approving credit scoring models.  

Watt, while somewhat soft pedaling any need for an immediate change in credit scoring models, acknowledges the importance of encouraging competition in that area.  He also worries that, in their attempt to win customers, providers of the models might "engage in a race to the bottom," i.e. compete to provide the highest scores. Goodman has a slightly different view, saying borrowers with a blemish on their record need to know why, and competition encourages this greater transparency. While there could be a tendency to do what Watt fears, she says, that practice can be curbed by requiring that each lender choose to use only one provider.

She also argues that competition can encourage faster incorporation of new information, and use of more available information, and thus provide more accurate individual scores.  One of the most promising avenues to do this is alternative data sources. Both utility and rent payment histories are important indicators of ability to pay for borrowers lacking traditional credit.  While only a small fraction of credit records includes information on positive rent and utility payment histories, the files contain negative information in those areas because delinquent debt ends up in collection.

Watt, however, notes both GSEs have recently been allowed to purchase loans processed through their automated underwriting systems that require the lender to certify a borrower's repayment history using nontraditional forms of credit, including rents and utilities.  

Goodman concludes that the current scoring models used by the GSEs and lenders who sell to them need to change. "The updated models have already been developed; it's time to conclude the ongoing studies and modernize the system. Incorporating newer models into the mortgage origination process would allow the market to serve a greater number of creditworthy borrowers seeking to purchase a home."

However, As MND reported yesterday, Watt explained why this isn't going to happen; not immediately at least.  In addition to saying the GSEs are already using the same or even greater levels of credit data than used by credit scoring, he said, "It now appears that it would be a serious mistake to change credit scoring models before the Common Securitization Platform is operational and the GSEs implement the Single Security, probably in mid-2019."  He added, "It also seems that, regardless of the credit score model decision, the short-term impact on access to credit will not be nearly as significant as first thought," citing again the broader application already used by the GSEs.