Bank of America puts the GSEs' internal scores ahead of external ones at predicting delinquency

The point gaps only mean something once you know how wide each scale is. Fannie's score runs 400 to 950 and Freddie's 200 to 1,000, against 300 to 850 for Classic FICO and for the newer external models the enterprises are adopting ^[9][6]. Fannie's band is 550 points wide, identical to FICO's and shifted up 100 at each end; Freddie's is 800 points, about 1.45 times FICO's width ^[1][2]. Bank of America's researchers wrote that "Due to their wider ranges and higher upper limits, both agency scores run higher than Classic FICO" ^[11]. Fannie's runs 10 to 20 points above it on average, Freddie's 20 to 30 ^[12][13]. Rescale Freddie's premium proportionally and it is worth roughly 14 to 21 points of FICO width ^[3]. The report does not describe how the scales map onto each other.

Freddie's score also runs 10 to 15 points above FICO 10T and VantageScore 4.0 ^[13]. Subtract that from its 20 to 30 point lead over Classic FICO and VS4 lands somewhere between 5 and 20 points above the traditional measure ^[4]. The enterprises have begun buying VS4 loans, and they want the VS4 number about 20 points above Classic FICO before the borrower gets the same pricing to start ^[14]. That opening requirement sits at the top of the implied band, up to 15 points more conservative than Freddie's own numbers suggest ^[5].

Who can act on the finding depends on when the number arrives. External scores get pulled at application, and the borrower and the lender see them before the file goes to the enterprises and before it is disclosed to investors ^[7]. The agency scores come out of automated underwriting; the enterprises traditionally see them at origination, and broader visibility comes through mortgage-backed securities disclosures ^[8].

Bill Pulte, the FHFA director who oversees the GSEs, said in a social media post that he welcomed the study's finding that the enterprises' scores "overall serve as better indicators of borrower delinquency risk" than some other measures ^[3]. Jeana Curro, Chris Flanagan, Ge Chu and Ko-Hsiang Kao wrote that "We think the providing of the agency score is a timely, robust addition to credit modeling, offering a consistent alternative" ^[2][1]. The report treats it as an addition. On prepayments, the researchers wrote that the score "would do little to anticipate faster (or potentially slower) voluntary speeds and convexity changes" ^[17].

On the question the industry has been arguing over, the researchers wrote that "The industry concern that lender choice among FICO and VS4 would meaningfully understate delinquency risk is not unfounded, however we find that the agency score largely helps to avoid this distortion" ^[16]. I think that is worth more to a bond buyer than to an originator, because the investor gets one yardstick across every lender's score choice while the underwriter still prices off whichever external score was pulled at application. Two things would change that view. The agency scores focus on mortgage delinquencies while external scores take in the borrower's broader credit profile ^[15], so the narrower measure cannot carry underwriting decisions that turn on non-mortgage credit. And Bank of America said its visibility into the scores is limited, and that differences between Fannie's and Freddie's measures complicate comparison ^[5][18].