The Hidden Tradeoffs of Lender Choice

At first glance, “lender choice” sounds like an unambiguously positive development. More choice typically implies more competition resulting in better pricing and improved outcomes for borrowers. In the context of mortgage credit, however, the reality is more complex. What appears to be a straightforward modernization of credit scoring may, in practice, introduce inefficiencies, shift risk in unintended ways, and ultimately increase borrower costs.

For more than two decades, the mortgage industry relied on a single credit score for evaluating borrower creditworthiness: FICO Classic. While other types of consumer credit have adopted updated credit scoring models, incorporating richer datasets and more dynamic methodologies, mortgage lending remained reliant on FICO Classic for approval and pricing. While a strong model, it was built on data and model methodologies that were limited to the data and compute capacity available when the score was developed over 25 years ago.

FHFA and the GSEs began a formal credit score modernization effort following the enactment of the Credit Score Competition Act in 2018. Over time, newer models including VantageScore 4.0 and FICO 10T were developed, incorporating trended data, expanded payment histories, and more modern analytic techniques. These models have generally been shown by their developers, and by the assessment during the modernization process, as offering improved predictive power relative to the legacy FICO Classic score.

The introduction of these models into mortgage underwriting was meant to reflect that progress. But the mechanism chosen to implement them (what is now referred to as “lender choice”) has fundamentally changed how these models function in practice. Lender choice allows originators to decide which credit score to use when delivering loans to the secondary market. On paper, this creates flexibility and encourages competition. For mortgage credit risk investors (e.g., mortgage insurance companies, Freddie Mac, Fannie Mae, and CRT investors), lender choice introduces a behavioral dynamic that impacts the relationship between borrower default rates and credit scores.

Any underwriting system that uses a scoring model to rank, approve and price risk assumes that the model is calibrated consistently with the application of the model. Lender choice invalidates this assumption because it modifies the way borrowers are assessed. During the mortgage origination process, Lender Choice can create incentives for lenders to select the approved score that yields the most favorable pricing for the borrower (i.e., lowest interest rate). A borrower who appears higher risk under one model may appear lower risk under another. Over time, this selection process shifts the composition of risk within each score band. The result is subtle but significant. Borrowers who would have been classified as higher risk under one system migrate into lower-risk categories under another. Meanwhile, truly high-risk borrowers remain concentrated at the lower end.

Empirically, this shows up in two ways. First, Milliman analysis indicates that default rates within the same nominal credit bands increase by approximately 30 percent. This is driven by the fact that those credit bands now contain a different mix of borrowers. Second, absent offsetting adjustments, Milliman analysis indicates that average risk-based pricing metrics, such as loan-level price adjustments (LLPAs), may decline, reflecting the apparent ‘mix-shift’ improvement in borrower profiles. Put another way, risk is rising, but pricing signals suggest the opposite. From a system perspective, that disconnect has consequences. Investors, insurers, and guarantors rely on stable, predictable relationships between credit scores and performance. When those relationships weaken, uncertainty increases. And in financial markets, uncertainty is always priced in.

That pricing response is unlikely to be immediate or uniform, but over time it becomes unavoidable. Over time, market participants may respond. Investors may demand higher required returns, Insurers may revise premiums, GSEs may recalibrate pricing grids, and market participants may demand enhanced data requirements. Each participant in the value chain responds rationally to the same signal: the underlying risk has become harder to measure.

And as is almost always the case in mortgage finance, those costs do not disappear. Ultimately, they will likely be passed through to the borrower in some way.

The irony is that a policy initiative partly motivated by a desire to reduce costs and expand access to credit may under its current structure, do the opposite. Lenders, seeking to remain competitive, may feel compelled to pull multiple credit scores for each borrower, increasing upfront costs. Investors, lacking historical data for this new multi-score environment, may price conservatively. Guarantors, facing higher realized defaults within score bands, may raise fees. On April 22, FHFA announced the Freddie Mac and Fannie Mae will begin accepting loans with credit scores other than FICO Classic; it was also announced that specific LLPA grids for these loans are being developed. At this time, it is not clear what the LLPA grids will be or if they will reflect the impact of lender choice.

Despite these hidden tradeoffs, modernizing credit assessment is not fundamentally misguided. On the contrary, updating credit models to incorporate enhanced data availability and scoring algorithms is a positive development for the mortgage market. The issue is with the uncertainty introduced by lender choice, which complicates the interpretation of what a given credit score means for a mortgage. That shift may seem minor, but it changes the incentives that drive behavior across the entire ecosystem. And in a market as interconnected as mortgage finance, small changes in incentives can produce outsized effects. Mortgage markets are not just collections of individual transactions; they are complex systems where origination, securitization, insurance, and investment are tightly linked. Changes at one point in the chain rarely stay contained. They propagate.

For policymakers, the challenge is not simply to evaluate whether a given change works in isolation, but to understand how it interacts with the system. Lender choice, in that sense, is less a discrete policy decision and more a structural shift; one that alters how risk is measured, priced, and ultimately borne. And that brings us back to a familiar truth. There is no such thing as a free lunch in mortgage finance. If risk increases, someone pays. The only question is when, and through which channel. In this case, the answer is likely to be gradual but clear. As the system adjusts, costs will find their way back to the borrower; not necessarily through a single, visible fee, but through a series of small, compounding adjustments across the value chain.

What began as an effort to introduce choice may ultimately serve as a reminder that in complex financial systems, more choice does not always mean better outcomes. Sometimes, it simply means more variables, and more to get right.

Jonathan Glowacki is a Principal and Consulting Actuary at Milliman

Milliman provides analytical and strategic solutions across the mortgage value chain, including research and insights on credit scoring, capital/risk, front-end mortgage origination, and servicing. Jonathan’s work focuses on helping market participants better understand and manage risks in mortgage credit, origination and servicing.