Automated Underwriting: Ten Years Later

Last year marked the tenth anniversary of Freddie Mac’s Loan Prospector, the first statistically-based automated underwriting (AU) system for home mortgages.  Shortly after Loan Prospector made its debut, Fannie Mae rolled out its own AU system, Desktop Underwriter.  By 2002, about two-thirds of all home mortgages were originated through Loan Prospector, Desktop Underwriter, or proprietary AU systems.

Automated underwriting has many benefits, not the least of which is faster loan decisions and lower origination costs.  Loan Prospector and Desktop Underwriter have also spurred greater flexibility in mortgage terms and underwriting standards.  Previously, under manual underwriting, any major negative often resulted in a “knock-out,” regardless of the applicant’s other strengths.  In contrast, the GSEs’ AU systems weight negative and compensating factors for risk and then offset the negative factors with the compensating factors as appropriate. The resulting flexibility has been especially important for minorities and lower-income borrowers, many of whom could not qualify for home loans under older, stricter manual underwriting standards.

So is AU the silver bullet for access to credit?  In our view, not yet. Although the GSEs have successfully used AU to expand credit to minority and lower-income borrowers, this is not necessarily the case with other proprietary AU systems.  One study of a bank’s in-house AU system discovered, for instance, that the system produced higher denial rates for low- and moderate-income borrowers than manual underwriting or relying on credit scores alone.

In addition, all AU systems suffer from serious omitted variable bias. AU only analyzes an applicant’s use of conventional credit, not fringe credit from pawnshops and payday lenders, and does not take rent and utility payments into account.  Similarly, AU cannot analyze loan applicants who lack or have only sparse credit histories.  These dynamics bar many minority and lower-income borrowers -- and numerous subprime borrowers -- from benefiting from AU.  Typically, B and C subprime loans are still manually underwritten.

Finally, AU is only as good as its inputs.  If an underwriter inputs wrong data – whether by ignoring red flags, not verifying data, or plain old fraud – then AU’s recommendations are invalid.  The rising incidence of mortgage fraud and sloppy underwriting makes this a real concern.  As they say in techspeak, “garbage in, garbage out.”