According to a report issued by Moody's Investors Service last week, mortgage credit scoring systems could have a mixed impact on the credit ratings, underwriting and servicing of mortgage-backed securities.
Mortgage scoring uses statistical or quantitative models for making mortgage lending and servicing decisions. It helps automate the assessment of credit risk and hence becomes a key component in any practical automated underwriting system.
According to David Zhai, vice president and senior analyst at Moody's, and author of the report, mortgage scoring "promises to improve profitability, consistency, objectivity, efficiency and competitiveness in mortgage underwriting and servicing processes."
However, Zhai says that the approach can also have pitfalls, based on the limits and biases in many existing scoring systems and misinterpretations of scores.
The report states that industry-wide competitive pressure compels lenders to pass at least some of the savings from underwriting cost reduction on to consumers through lower upfront costs and better loan pricing. Additionally, the objectivity of a scoring system helps assure the underwriting process is blind to an applicant's race, color, national origin, gender, religion or marital status, and is otherwise fair and consistent.
Moreover, scoring technology also reduces the time for credit assessment, making it possible for homebuyers to secure financing more quickly and easily than ever before.
"Scoring models cannot succeed without sustained quality control and constant vigilence against fraud," Zhai said. "And no matter how much effort and resources have been invested in creating a scoring system, all mathematical or predictive models have limits."
Some typical problems in scoring models arise from building models based on mortgage performance datat from good economic times. Using that data could lure underwritiers into making overly aggressive loans that result in significant losses when the economy inevitably turns downward, according to Zhai.
Other problems can range from model biases caused by using a narrowly defined universe of data, either in terms of loan characteristics or the time frame of performance data, to reduction in predictive power for bad loans when using data only from approved loans.
According to Zhai, taking a system developed for another use - such as evaluating consumer credit loans - and then applying it to mortgage loans may give a false read on defaults. That is because most consumer loans are for less than two or three years while most mortgage defaults occur in years three to seven of a loan.
Additionally, application fraud is reported to be on the rise as automated underwriting becomes more popular, transparent and publicized. Examples of application fraud include identity theft, income fraud, occupancy fraud and appraisal fraud. Some key market players, including Fannie Mae, have expressed concerns that scoring models may invite fraud if their inner workings are exposed to view.
The report went on to briefly review several mortgage and credit scores, including ACUScore by United Guaranty, FICO by Fair, Isaac and Desktop Underwriting by Fannie Mae. It is apparent from the descriptions of different mortgage scoring methods that some models are more complex than others. Many have unique strengths for specialized applications in mortgage underwriting and servicing.
