Quality control (QC) experts from Fannie Mae and Freddie Mac shared insights about the trends they are seeing among performing and non-performing loans while speaking at the Mortgage Bankers Association’s (MBA) 2025 Compliance and Risk Management Conference.

MBA Chief Economist Mike Fratantoni facilitated the discussion, which featured Fannie Mae Senior Director, Loan Quality Compliance Duane Gilkison and Freddie Mac Senior Director, Quality Control Stephen Nally.

More than 99 percent of loans reviewed by Fannie Mae in the first two quarters of 2025 were performing loans, they noted. The two companies have taken slightly different approaches to strengthen their early defect detection capabilities, improve data analytics and utilize artificial intelligence (AI) and machine learning (ML) to help prevent both fraud and performance deterioration.

Random, targeted sampling rates

Fannie Mae and Freddie Mac have similar core quality control objectives with some nuanced differences in sampling approaches. Both use random and targeted sampling with a focus on performing loan samples — stratified based on their largest customers — and track defect rates across multiple categories.

Nally explained Freddie Mac’s method emphasizes performing loan (PL) samples over non-performing loan (NPL) samples, particularly when assessing emerging new products or policy changes. The company recently expanded its PL repurchase alternative pilot program to include more than 500 sellers.

“Our PL targeted sample is used to evaluate the effectiveness of a new product or policy change,” Nally said. “We also use this sample when we see emerging issues and need to take a deeper look and the feedback from this sample is used by internal groups such as credit policy to adjust policies if necessary.”

Gilkison described Fannie Mae’s sampling methodology as a bit more granular, relying on supplemental random samples to ensure coverage across the entire delivering lender base, which helps when assessing loan quality. Random sampling provides a broad measure of portfolio quality, while targeted samples focus on loans flagged by data analytics as higher risk, he said.

“For our random sample, we sample it within the three weeks of the acquisition month closing out, and we have our statistically valid defect rate probably before you do,” Gilkison told Fratantoni, noting Fannie Mae had already begun sampling work for its September portfolio. The QC team will then stratify that data into subgroups to provide a detailed perspective on which quality issues are having the most notable effects on the overall defect rate among loans purchased by the enterprise.

Targeted samples utilized by Fannie Mae are based on data indicators pointing to potential quality issues, he explained.

“When the defect rate moves either up or down, people always ask questions,” Gilkison said. In discussing the random sample he stated, “If you don’t have some more granular cuts of that data, at least for the size of the book we have, it’s hard for us to really pinpoint what can be driving defect rates. So, for our largest customers, we do have a specific, statistically balanced slice. Outside of those larger customers, we do other customer groupings and bigger buckets so that we have some visibility into what drives that defect rate.”

He credited Candace Kubida, Fannie Mae’s senior director of loan quality, for her work disseminating articles detailing the enterprise’s findings on the most prevalent types of defects affecting loan performance, repurchase data and emerging risks.

Top defect drivers

Among the most prevalent defect categories tracked by the GSEs over the first two quarters of the year, half were income-related. These included incorrect rental income/loss calculations and incorrect base income calculations.

However, the top source of loan defects was one that has also been the most heavily discussed in the news cycle — misrepresentation of occupancy.

Occupancy discrepancies are typically not the result of attempted fraud and the best way for a borrower or a lender to prove that is through documentation.

“Our defects are misrepresentation defects, meaning the facts have been misrepresented,” Gilkison explained. “We look at those on a case-by-case basis. Sometimes we’ll see where a borrower closed on a loan and got relocated for work. That happens. And you have documentation from your employer that you got relocated. There would be documentation for a change of circumstances like that. We also have people who decide after closing that they don’t like the house. Well, that’s not a change in circumstances.”

The bottom line for the GSEs when looking into defects is determining the right outcome, he said. Typical fraud scenarios include:

• Borrowers claiming owner occupancy while intending to rent the property
• Misrepresenting reasons for property changes
• Listing properties for rent shortly after closing

The GSEs’ strategies for detecting these misrepresentations rely on a mix of traditional verification practices and advanced analytics, including reviewing transactions with borrowers substantially buying-down in price, checking online rental listings, analyzing inconsistencies in property data, validating insurance policies and examining public property records.

The majority of defects involving occupancy misrepresentation occur when borrowers claim owner-occupancy status for a property they intend to use as a rental property. These incidents become apparent when these buyers misrepresent the reasons behind property changes or list these homes for rent soon after closing.

Both organizations have emphasized to lenders they should shift their QC insights earlier in the loan origination process, using data and technology to identify potential issues before loan closing. Concurrently, they are both also exploring ways to apply new technologies to improve loan quality, reduce fraud and create more efficient processes.

Recognizing the potential to make their defect detection processes more efficient, Freddie Mac is investing in technological modernization by upgrading its core QC systems, including its AI-powered tool for conducting property condition assessments. Gilkison described Fannie Mae’s work in applying AI to its work to standardize repurchase letter writing, while also using machine learning to analyze loan performance patterns and develop tools that can flag potential issues faster.