A new credit scoring model — expected to roll out in fall 2017 — aims to more accurately measure credit risk by using more historical data and machine-learning techniques while culling less reliable information.
VantageScore Solutions announced the release of the fourth version of its credit scoring model, to be used by the three national credit bureaus. (It may seem like these credit scoring models don’t change much, but they do update from time to time. You can find out 13 ways credit scores have changed in the past 20 years here.)
VantageScore 4.0 improves on its predecessor in three main ways, said Sarah Davies, senior vice president of product management and analytics for VantageScore. First, it looks at a consumer’s credit behavior over time, incorporating more of what’s known as “trended data.”
For example, the score takes into account how a consumer’s credit balance has changed over a period of months, rather than taking a single snapshot in time. This has made the score upwards of 20% more predictive than 3.0 for customers with good credit, Davies said.
The new model also excludes a lot of public record information, especially liens and judgments. In many cases, Davies said, this information was difficult to accurately link to individual consumers.
“In all likelihood, almost all civil judgments will be removed from credit files and a substantial portion of tax liens will be removed from credit files,” Davies said.
With this new model, medical collections won’t be reported on credit files until after six months have passed. That’s because there is often confusion as to whether the consumer or insurer is responsible for the payment, Davies said.
The third big update is the use of machine-learning techniques to help score consumers with thin credit files. VantageScore used large data-processing platforms to examine thousands and thousands of combinations of consumer behaviors to identify which ones were associated with people paying their bills on time.
Despite the high-tech method, this led to some intuitive conclusions, Davies said. For example, for consumers with big collections accounts, VantageScore was able to parse out that those who are looking for credit are higher risks than those who aren’t.
It seems obvious, but a human may not have identified this as a risk factor. Davies said the technique has made VantageScore 4.0 about 17% more predictive than its predecessor for people who haven’t used credit in the last six months and 30% more predictive for people who don’t have activity on any accounts, just collections.