Fannie Mae’s repurchase reform goal: end buybacks with AI

Fannie Mae and Freddie Mac are doing more to address what have been expensive repurchase concerns, with the former noting potential for artificial intelligence to end them long-term and the latter working to clear roadblocks to current technology use that minimizes them.

“Our North Star is to get out of repurchases altogether,” Devang Doshi, senior vice president of capital markets at Fannie, told the Mortgage Bankers Association’s Secondary and Capital Markets Conference in New York on Tuesday, noting that it’s a goal shared by colleagues in the market.

Doshi said Fannie is looking into how the application of generative artificial-intelligence, machine learning and other technology could be used to provide loan data integrity upfront, removing the need to call upon lenders to buy back mortgages due to flaws.

His comments suggest emerging technology makes the GSEs’ longstanding zero-defect loan aim more plausible.

Some tech experts attending the event agreed that repurchases could eventually end if AI, which allows for ingestion and analysis of much larger datasets than in the past, can be applied to digital verifications of information in compliant ways, and companies adopt it.

“I absolutely think it’s possible if there is full transparency and validity of the data at the time of sale,” said Cade Thompson, co-president of Rocktop Technologies, a fixed-income automation and consulting firm active in the mortgage industry, in an interview.

But interim steps like a fee-based repurchase alternative Freddie’s testing through a pilot program may need to come first.

Both government-sponsored enterprises, which buy a significant number of the mortgages in the market, have been advocating for greater data integrity for some time by offering or approving digital tools aimed at efficiently validating the information against existing records. 

But there have been questions about the costs.

Consumer Financial Protection Bureau Director Rohit Chopra in a speech about credit data providers and an industry group have questioned the fees associated with a GSE approved data verification tool in this area offered by Equifax.

In response to past claims like Chopra’s assertion that “Equifax’s market dominance has given it pricing power,” the pioneer in the space has pointed to current competitors ranging from fintechs to free requests for consumer paystubs. The GSEs also have expanded approved alternatives.

Also, Freddie found in a recent study that generally use of its data validation tools offer significant savings in addition to a 40% reduction in defects found in previous research.

“There’s a 14% lower cost than folks who are not using our tools,” Sonu Mittal, senior vice president and head of single-family acquisitions at Freddie Mac, told conference attendees, highlighting a finding in retail lending research the GSE released at the conference.

Although the industry, GSEs and rate environment have reduced buybacks somewhat as a concern from a year ago, savings in cost and time found in the study should be attractive in a market where mortgage bankers recently absorbed a record annual loss, the study suggests.

In addition to lowering the average retail mortgage cost by roughly 14% or $1,500 per loan, each of various types of digital data validations could save from around 2 to 12 hours of processing time or an additional $28 to $179 per component, according to Freddie’s study.

However, that technology has been underutilized and Freddie’s has been examining the causes.

In its research, one thing Freddie has found is that these validations may get redundantly applied at the same time as traditional processes that don’t account for the new method, undermining the savings and discouraging usage, said Kevin Kauffman, the GSE’s head of client engagement.

He cited the example of a loan in which the digital verification that should have sufficed was utilized, but a traditional prompt in the origination system resulted in redundant request for pay stub information. Even if this does not involve a fee, it adds to processing time.

Freddie Mac’s work with Intercontinental Exchange’s mortgage technology division on data quality initiatives could help address such issues.

However, there also are hurdles at the lender level that can stand in the way of more upfront data integrity and cost savings, he said, noting that staff at mortgage companies also need to help guide borrowers in new processes that often are consumer permissioned, but may not.

“You can’t just send a borrower the application, you have to coach them through certain things,” Kauffman said, noting that he finds that loan officer buy-in to this varies and sometimes there is LO resistance standing in the way of lender reduction of buyback risk and savings.

Freddie Mac’s findings dovetailed with those of the newly-merged Stratmor Group and Teraverde, which also suggest resistance to a new dynamic in origination and loan sales is an impediment to the most cost-effective use of this and other technology.

Rethinking the role technology plays in customer satisfaction and repeat business or referrals when analyzing costs and revenue may be part of the answer, said Garth Graham, a senior partner at Stratmor, in an interview.

Consumer surveys Stratmor has done show when it comes to the question of lender choice, “90% of it is driven by some level of customer experience,” Graham said, noting that this extends throughout the organization from sales to broader lender operations.

Although Realtor referrals are the largest piece of that (26%), which is interesting in light of changes that industry may be undergoing, the rest is in the lender’s control. For customer loyalty and advocacy, process accounts for 83% and the LO relationship 17%, Stratmor found.

Metrics around things like customer satisfaction scores and efficiency some consumer finance companies have used to analyze performance and determine compensation for fulfillment positions can be extended to loan officers to motivate change, he said.