Lenders wrestle with the nuances of modern credit score pric

Mortgage lenders are currently wrestling with a high-stakes puzzle: how to accurately compare the new credit score models hitting the market. The differences are significant and carry major implications for risk assessment, Not to be left behind, U.S. Department of Housing and Urban Development (HUD) Secretary Guild Mortgage has already started to compare how the same loan scores across all three models, but it’s in the “very early stages of collecting this data,” according to David Battany, the company’s executive vice president for capital markets.

“When you see a delta between two models of 40 or 80 points, that’s pretty significant,” Battany said this week during a session on the new credit score models at the Tricky calibration

FICO and VantageScore themselves warned about the nuances between their models and the potential headaches of calibration.

“I wouldn’t underscore it and make it sound too easy to do the calibration,” said Ethan Dornhelm, head of scores analytics at FICO. “Certainly, calibrations are possible, but we do see that there can be drift over time, and given that the two algorithms that are the modernized credit scores are different, they could drift slightly differently. So, it will be a case of not just calibrating one time and being done with it, but rather careful and close monitoring over time.”

VantageScore recommends probability-of-default mapping, saying that translation tables are straightforward.

“What we recommend doing is to basing it on the probability default, so you have the similar expected outcomes when you’re looking at converting the scores, and we’re about to provide some more data on that,” said Rikard Bandebo, chief strategy officer and chief economist at VantageScore.

Most lender systems are built around a single credit-score field. Adding a second introduces major operational and policy questions. A concern is with cherry-picking, since lenders might simply submit whichever score makes the loan look more affordable for the borrower and ignore the actual probability of default.

“With regards to Bob Johnson, head of originations at Newrez, explained that Freddie Mac approached the company to see if it could “test the plumbing.” Newrez was “able to make those deliveries and fully test out the system,” he said.

But the secondary market was not fully tested, sources said. The first multilender GSE securitization containing VantageScore-underwritten loans totaled just under $8 million within an $11 billion pool, according to Dornhelm. He added that FICO reported one single-lender securitization under FICO 10T, with more expected in the home equity line of credit (What makes them different is that both FICO 10T and VantageScore 4.0 use time-series balance, payment and utilization data rather than a single point-in-time snapshot. Both new scores factor in rental payment history — although less than 5% of files currently contain it, which represents a major growth area for financial inclusion. They are also built on more recent data and better reflect modern consumer behaviors.

But points of contention remain. While FICO 10T builds bespoke models for each bureau, saying that it maximized their unique data, VantageScore 4.0 uses one algorithm across all three bureaus for score consistency.