A change coming to credit reports and scores this summer may inspire an overhaul in the data and technology banks use for credit modeling.
Come July, TransUnion, Experian and Equifax will no longer include information about tax liens and civil judgments on a consumer’s record if the data doesn't include the person’s name, address, Social Security number and date of birth. Many liens and most judgments don't include all that data, in part because Social Security numbers are often redacted for security reasons. Some consumers could see their credit scores rise with the removal of such black marks.
The change is the result of a settlement the three major credit bureaus made in 2016 with 31 state attorneys general over alleged problems with credit reporting accuracy. The state governments would prefer credit bureaus not include deeply disparaging information in credit reports unless they’re sure it’s true. LexisNexis Risk Solutions has estimated that 50% of public-record information about tax liens and 96% of information about civil judgments do not include a full or redacted Social Security number and will not meet the new credit bureau requirements resulting from their settlement with the 31 states. LexisNexis Risk Solutions sources tax lien and civil judgment information from government public record sources and provides it under contract with each of the three credit bureaus. It also provides the same type of data directly to lenders.
Whether this change is a good thing depends on whom you ask.
Some observers note that tax lien and civil judgment information is sometimes attached to the wrong consumer’s file due to a lack of identity information.
“The bureaus are saying they don’t want to unnecessarily penalize someone with this information if it’s not the right consumer,” said Sarah Davies, senior vice president of analytics, product management and research at VantageScore. “So this is a story about correcting consumer scores.”
In a study last year, VantageScore found that 11% of consumers had either a lien or a judgment on their credit file. But “there’s still this situation where they could be misidentified,” she said. “Removing this stuff and making it accurate is the right thing to do.”
Others argue it’s important for lenders to know if consumers have had a lien on their taxes or a civil judgment against them, because their risk of defaulting on a new loan is much higher.
“It's a bad idea to remove anything from a credit report that's predictive of future credit risk and is accurate,” said John Ulzheimer, a credit reporting and credit scoring expert. “Liens and judgments have been a part of credit reports for decades. Can they end up on the wrong consumer's credit report? Yes. Is it such an overwhelming and common problem to eliminate almost all judgments and half of the liens from credit reports? No.”
This move dilutes the value of credit reports and credit scores, Ulzheimer said. “There's also the very real potential for consumers who have these pubic records will have higher scores after July 1,” he said. “The real question is does the deletion of these public records equate to better credit risk? The answer is clearly not.”
And some, for instance those who deal only in prime credit, see little effect.
“It is not a significant issue for Huntington with relatively limited impact to our business,” said Tim Barber, executive vice president of credit risk management at the $100 billion-asset Huntington Bancshares in Columbus, Ohio. “However, it will likely create issues for subprime originators.”
Regardless of how they feel about the change, banks will have to do some rethinking of the credit scores they use in their underwriting models starting in July, simply because some information they used to rely on will no longer be included.
In the bigger picture, it may be time to do this anyway, as the world is changing and new forms of alternative data and machine learning methods of analyzing that data have emerged (which come with their own risks and downsides, of course) that online lending competitors have been taking full advantage of.
“It is encouraging to see that through innovation, credit scoring models are becoming more predictive and inclusive,” said Debra Still, president and CEO of Pulte Mortgage. She sees benefit in the recent availability of trended data and says the real estate finance community should gravitate toward alternative credit scoring models.
Lenders will have to adapt
At the online lender Enova, Chief Analytics Officer Joe DeCosmo said he views the removal of public records from credit files as an important development because it will affect some of the credit scores the company uses.
“We're going to get some of the new, updated scores and try them on samples of our data, so we'll be able to see the difference,” he said. “We’re testing how the score we use changes and how we have to change our models because of it. And if there are other sources of this data, is it worth acquiring them separately?”
Because Enova already looks at a lot of data in credit reports beyond the standard credit scores, as well as alternative sources of data, DeCosmo said he doesn’t think the change will affect Enova’s business as much as some other businesses that are more reliant on third-party-produced scoring.
“At many banks and mortgage lenders, standard credit scores are their primary criterion,” DeCosmo said. “For us, it's just one of many different variables and data sources we use in our models.”
How credit scores will change
FICO, which says its credit scores are used by 90% of U.S. lenders, says it doesn’t see the need to change its credit model to accommodate the loss of public records.
The company projects that just less than 11 million consumers will see a score increase of less than 20 points.
The number of consumers who will see a bigger bump in their score is small, FICO said, because most consumers who have a tax lien or judgment on their file still have other derogatory indicators such as collections or serious delinquencies on their credit file, which will remain after the public record information is removed.
“Roughly 0.35% of the total scorable population, or some 700,000 U.S. consumers, are projected to have a score increase of 40 or more points as a result of the updated public record retention policy,” said Ethan Dornhelm, vice president of scores and analytics at FICO.
VantageScore, the joint venture that is owned by the three credit bureaus and provides credit score data to 2,400 banks and other companies, will announce a new, updated credit score model on Monday.
In its 4.0 version of VantageScore, the company is reacting to the absence of tax lien and civil judgment data by adding “trended data” it gets from all three credit bureaus, as well as adding machine learning technology.
Trended data is a monthly snapshot of certain pieces of information presented in a way that shows how a consumer’s behavior changes over time — in increased or decreased credit card use, for instance. An increase in use is interpreted as higher risk and will drive the score down.
“This trended data is teasing out a little bit of the nuance and allowing us to get a better read on the consumer in terms of their likely behavior,” Davies said.
VantageScore’s tests have shown a 20% improvement in a score’s ability to predict the performance of prime consumers through the use of trended data.
“You don’t often see this kind of lift,” Davies said.
VantageScore has tested credit score models with and without the tax lien and civil judgment data.
“We found that absolutely you could swap in other information, and the model would be just as accurate at predicting risk as the old model based on public records,” Davies said.
Specifically, VantageScore has replaced the public record data with the number of high-credit-limit credit cards in consumers' credit files. Where the record of a tax lien or civil judgment might be several years old, the credit card data is forward looking and can therefore give a slightly better indication of behavior going forward, Davies said.
VantageScore is also adding machine learning to the latest version of its model. It’s using a technique called random forest tree to randomly select different behaviors from a consumer’s credit file and see how that combination of behaviors affects creditworthiness. The machine learning software can calculate the effects of different combinations tens of thousands of times to see what sticks — in other words, which combinations of behaviors yield a lot of people who ultimately default. In one test set, machine learning discovered that consumers who have less than $1,000 in a medical collection account that’s more than 36 months old and they’ve applied for credit more than three times recently, have a 32% default rate.
“Because you’re doing it so many times with so many combinations of behaviors, the cream rises to the top and you end up identifying a set of information that’s not intuitive to the human brain,” Davies said.
Editor at Large Penny Crosman welcomes feedback at firstname.lastname@example.org.