In September 2026, Fannie Mae and Freddie Mac each published their historical credit score data for the first time. The government-sponsored enterprises (GSEs) then began releasing their internal scores on new originations in October. Both GSEs describe the efforts as ways to provide more insight into their internal credit risk assessment processes as they modernize their credit scoring processes.
The fact that lenders can now deliver either Classic FICO scores or VantageScore 4.0 scores could lead to credit score shopping, in which lenders deliver whichever score minimizes the risk-based guarantee fees (i.e., the loan-level pricing adjustments, or LLPAs) the GSEs collect. We argue that this would reduce guarantee fees and, as such, the GSEs’ profitability, and it would make credit risk transfer (CRT) deals less transparent for investors, potentially raising the cost of these deals to the GSEs.
GSE scores and commercially available scores do different jobs. The GSE scores are internal risk tools, used in Fannie Mae’s Desktop Underwriter and Freddie Mac’s Loan Product Advisor, and neither borrowers nor lenders see them when a borrower applies. The commercial scores are visible to lenders, mortgage insurers, and investors, so borrowers and lenders know how a score will affect pricing before the borrower applies.
The GSEs’ recent transparency efforts mitigate our concerns about the impact of credit score shopping on CRT deals, as investors will have the GSEs’ internal scores to calibrate their risk assessment. But because the GSE scores are not available at the time of application and because LLPAs are based on the commercial credit score at the time of application, lenders can still deliver whichever score puts the loan in a lower LLPA tier. As such, we remain concerned that the risk-based guarantee fees the GSEs collect will be lower.
In this article, we compare commercially available scores from Classic FICO, FICO 10T, and VantageScore 4.0 with each GSE’s disclosed loan-level score. We show that both GSEs’ scores concentrate defaults among their lowest-scoring loans to a greater extent than any of these commercially available scores. The gap is widest for first-time homebuyers and for purchase loans.
How the GSEs’ scores differ from the commercial scores
The commercially available scores used in the mortgage market are designed to predict delinquency on any type of credit account. In contrast, the GSEs’ own scores are built to predict mortgage delinquency and are estimated on each GSE’s own historical loans.
Both GSEs describe their scores as being built from credit data. Fannie Mae says its score is generated from credit repository data, while Freddie Mac’s score is based on traditional and trended credit data and, when available, rental payment data. Each score is one input in Desktop Underwriter or Loan Product Advisor, rather than the full underwriting decision. The GSEs’ own scores are generated at the time of loan approval, while commercial scores are pulled at loan application.
The GSEs use their own scales, typically 400 to 950 for Fannie Score and 240 to 1,000 for Freddie Score. As such, we compare the scores by how they rank loans, not by the scores themselves.
We find that the GSEs’ scores rank borrowers similarly, but not identically, to the commercially available scores. When each loan is placed in a decile under each score, 27 to 31 percent of Fannie Mae loans fall in the same decile under Fannie Score as under a commercial score. For Freddie Mac, the share is 20 to 23 percent.
Fannie Mae and Freddie Mac scores agree most on the riskiest borrowers: About half the loans in Classic FICO’s lowest decile are also in the GSEs’ score’s lowest decile. Both scores line up more closely with the two trended-data models than with Classic FICO, consistent with the GSEs’ use of trended credit data.
How the GSEs’ scores capture defaults
Using the same methodology we used in our recent brief, we sort loans from the lowest score to the highest score, split them into 10 equal groups, and measure what share of all loans that ever became 90 or more days delinquent falls in the lowest-scoring groups. A score with no information would place 10 percent of defaults in the lowest-scoring 10 percent of loans; a higher share in the lower-scoring buckets is a measure of the model’s predictive power. The results cover loans the GSEs acquired, which had already passed GSE underwriting. The measure does not capture effects on access to credit, including for borrowers who can be scored under some models but not others.
The commercial scores rank similarly to what we found in our previous analysis. In the lowest-scoring 10 percent of loans, FICO 10T captures the most defaults (32.5 percent at Fannie Mae and 31.6 percent at Freddie Mac), followed by VantageScore 4.0 and Classic FICO (29.0 to 31.1 percent).
In comparison, the lowest 10 percent of the GSEs’ scores capture more defaults: 37.3 percent of defaults under Fannie Score and 37.4 percent under Freddie Score. This advantage may reflect that each score was estimated based on mortgage loans only—in particular, mortgage loans similar to those evaluated here.
Results by homebuyer type and loan purpose
Next, we ran the same analysis by borrower groups, looking only at loans in the lowest 10 percent in terms of score.
We find that first-time buyers tend to have thinner credit files. For them, the lowest-scoring decile under the commercial scores holds a smaller share of defaults than for repeat buyers at both GSEs.
The GSE scores change less between the two groups (35.7 and 36.3 percent at Fannie Mae, 37.1 percent for both at Freddie Mac). In addition, the improved performance of the GSE scores versus the commercial scores is more pronounced for first-time buyers than for repeat buyers.
When we split the loans by purpose, we find that across all four scores, the models are less predictive for cash-out refinances than for purchase loans or for rate-term refinances.
Though the commercial models are more predictive for rate-term refinances than for purchase loans, the GSEs’ scores show little difference in prediction between the categories. For purchase loans, the share of defaults captured by the bottom decile is 7.9 percentage points higher under Fannie Score and 9.9 percentage points higher under Freddie Score than under Classic FICO.
Implications for the GSEs
Beginning in October 2026, the GSEs’ mortgage-backed securities disclosures have reported Fannie Score or Freddie Score, as applicable, and a pricing indicator identifying which commercial score was used to price each loan. The GSE score gives investors one consistent risk measure across loans priced on different commercial scores.
We believe this disclosure of the internal Fannie Mae and Freddie Mac scores is particularly important for investors in CRT securities. We had been concerned that credit score shopping (i.e., seeking the highest score to minimize LLPAs) would make it hard for CRT investors to evaluate the credit risk on their securities. If CRT investors were unsure to what extent the scores were shopped, they would likely assume the worst and price accordingly, which would then cost the GSEs more to sell off some of their credit risk. Now, with the GSEs’ disclosure of their own scores, investors have consistent benchmarks.
In contrast, this disclosure does not change pricing. LLPAs are still based on the commercial scores, which borrowers and lenders can see before applying. Lenders can deliver whichever score puts the loan in a lower LLPA tier, so credit score shopping could continue to reduce the risk-based guarantee fees the GSEs collect, reducing the GSEs’ profitability.
