FICO vs. VantageScore 4: What the GSE Credit Score Data Drop Actually Means for Lenders

Fannie and Freddie just released counterfactual data on 53 million loans. Greg Oliven walked the Risk & Roll panel through what it shows, and the answers were more nuanced than most people expected.

Credit score modernization has been in motion since 2018, and after years of slow rolling, the GSEs just dropped a significant piece of the puzzle: a loan-level counterfactual data set covering 53 million loans over 12 years, showing what would have happened if lenders had been using FICO 10T and VantageScore 4 all along. Greg Oliven from Polygon Research brought the data, the panel brought the questions, and the conversation got into the weeds on what this actually means for lenders, borrowers, and the three bureaus sitting at the center of all of it.

Featured voices: Greg Oliven (Polygon Research) · Dana Georgiou (Dunmore) · Bob Simpson (Daylight AML) · Nathan Knottingham (MLO Force, Host)

🎬 Watch the Full Episode: Greg shares his screen and walks through the data live.

https://www.youtube.com/watch?v=nb5CGtZ4xVc

What the GSE Data Drop Actually Is

This wasn't the GSEs flipping a switch on new credit score requirements. That rollout is still moving slowly, with only a handful of approved lenders currently using FICO 10T or VantageScore 4 in practice. What Fannie and Freddie released, in what Greg described as a clearly coordinated FHFA publication, is historical counterfactual data: what would the numbers have looked like across 53 million loans from 2013 through 2025 if both new models had been in use?

There are three parallel modernization tracks in motion right now: the move from tri-merge to bi-merge credit pulls, a methodology shift from middle/lower/lowest scoring to an average-average approach, and the introduction of new scoring engines alongside classic FICO. This data drop touches all three, and it's the most concrete look yet at how they interact.

How the Two Models Actually Compare

The headline finding is that FICO 10T and VantageScore 4 correlate strongly at about 0.86, but that number hides meaningful variation at the loan level. Forty-seven percent of loans differ by at least 20 points between the two models, and 17% differ by at least 40 points. Neither model consistently scores higher either: VantageScore 4 came in higher on just under 50% of loans, FICO 10T on 48%, with the remaining 1.47% tied.

Dana noted that even an 8-point difference could matter in practice, since LLPA pricing buckets mean a score bump at the right threshold translates directly to better pricing for the borrower. Nathan pushed back slightly, pointing out that higher scores don't always mean better pricing once LTV enters the picture, and that the relationship between score and rate isn't always linear depending on the loan structure.

"Competition, transparency, know your data. That's the punchline for today."— Greg Oliven, Polygon Research

The Methodology Shift and What It Does to Scores

Switching from the current middle/lower/lowest calculation to the proposed average-average approach, holding everything else constant, raised FICO 10T scores by an average of 8.18 points and VantageScore 4 by 6.14 points. A modest but real bump, and one that could move borrowers across pricing thresholds in meaningful numbers at scale.

The bi-merge variation data was also telling. When looking at all possible two-bureau combinations, the spread between the best and worst pair averaged 9.2 points for FICO 10T, with a 90th percentile spread of 18 points. VantageScore 4 came in tighter at 6.19 average and 14 points at the 90th percentile. Dana had expected a larger gap, and her reaction captured what a lot of people in the industry are probably feeling.

"Higher cost, not a really meaningful differentiator across the board. Why are we spending so much time and resource on this when there are so many other issues in housing that could take our attention?"— Dana Georgiou, Dunmore

The Cost Problem Nobody Predicted

One of the more frustrating threads in the episode: VantageScore was supposed to introduce competition that brought credit report costs down. It hasn't. Nathan reported seeing costs move in the wrong direction entirely, with VantageScore pricing rising to match FICO rather than undercutting it. The promise of a more affordable consumer product hasn't materialized, and Dana pointed out that it's the borrower absorbing those costs in the end.

Nathan traced the real cost pressure upstream to the three bureaus themselves: the scoring models are calculated from bureau data, and as long as Experian, Equifax, and TransUnion control that data and set their own prices, competition at the scoring engine level may not move the needle much for consumers.

Bob's Framework: It All Comes Down to Default Prediction

Bob Simpson reframed the whole conversation with the clearest possible lens: the only thing that actually matters here is whether the new models predict default rates more accurately than the old ones. If you know your default rate, you can price for it. If you don't, you're guessing. The historical lookback data Greg described is valuable precisely because it lets the industry test whether these models would have called it right.

"Anybody can make any amount of money if you know what your default rate is. All we're talking about here is predictability."— Bob Simpson, Daylight AML

Bob also raised the broader question of whether the three-bureau system itself makes sense in a competitive market. His argument: in most industries, competition produces a winner. In credit reporting, three players have coexisted for decades without any of them winning outright, which benefits shareholders more than consumers or lenders.

What This Means for Thin-File and First-Time Borrowers

Nathan raised the use case that arguably matters most: borrowers with thin or nonexistent credit files. He's working with a borrower in Iowa who intentionally avoided the credit system for years, eventually opening two cards and self-reporting utility trade lines to build a score. The promise of newer models like VantageScore 4, which appear to weight recent revolving credit behavior more heavily, is that borrowers like this get a fairer read. Whether that's playing out in practice is less clear, and Nathan's anecdotal read from loan officer groups is that results have been about 50/50, with VantageScore sometimes coming in lower than FICO on the same borrower.

Greg noted that the trending orientation of both new models, the T in FICO 10T stands for trending, is specifically designed to capture these kinds of borrowers more accurately than a classic point-in-time snapshot. The long-term proof will be in the performance data.

Key Takeaways

• The GSE data release is counterfactual, not a rollout. New scoring models are still trickling into live originations with a small number of approved lenders.

• FICO 10T and VantageScore 4 correlate strongly overall, but nearly half of all loans show at least a 20-point difference between the two models.

• Neither model consistently scores higher. The split is roughly even across 53 million loans.

• Switching to average-average methodology adds about 8 points to FICO 10T scores and 6 points to VantageScore 4 on average, enough to move some borrowers into better pricing buckets.

• Credit report costs have not come down with the introduction of VantageScore. Competition at the scoring engine level hasn't moved bureau pricing.

• The ultimate test is default prediction accuracy. That's what the historical data will tell us, and it's the only number that settles the debate.

• The bureau structure itself may be the bigger long-term question. As long as three entrenched players control the underlying data, competition at the model level has limits.

• 
  

Risk & Roll covers mortgage compliance, risk, and industry strategy every episode. Subscribe wherever you listen, and leave your feedback in the comments or at support@mloforce.com.