VantageScore Solutions, the company behind the VantageScore credit
scoring models, announced today that it has published its latest model
performance assessment of VantageScore 4.0, the company’s most recently
introduced consumer credit scoring model.
Data scientists at VantageScore Solutions examined the performance of
VantageScore 4.0 in three specific areas:
1. Predictiveness for both the mainstream consumer population
(i.e., those who are conventionally scored by generic scoring models)
and the approximately 40 million consumers who are considered “Newly
Scored” (i.e., those who can be scored with VantageScore 4.0 but fail to
meet the criteria required by conventional models).
2. Consistency of VantageScore 4.0, based on the credit scores
obtained from the three national credit reporting companies (CRCs) in
terms of the variance between the scores and predictive performance.
3. Statistical bias, if any, related to the ethnicity of various
consumer groups.
The model performance assessment, conducted in 2019, represents the
second annual assessment of VantageScore 4.0 (which had been launched in
April 2017). The model is the first and only tri-bureau credit scoring
model to incorporate trended credit data and leverage machine learning
for superior predictive insights and inclusivity.
Highlights of the 2019 model performance assessment include:
-
Across ALL major consumer-lending categories (Bankcard, Mortgage,
Auto, Installment), VantageScore 4.0 achieved a predictive performance
lift over a best-in-class benchmark CRC proprietary credit scoring
model for both originations and account management. -
VantageScore 4.0 continues to significantly outperform VantageScore
3.0 in Bankcard, Auto, and Mortgage Originations. -
There is no statistically significant difference in the default rates
observed within each VantageScore 4.0 score band for Mainstream
consumers as compared with the Newly Scored consumers (who often are
referred to as “credit invisibles” under conventional scoring models). -
Among all models introduced by VantageScore, VantageScore 4.0 provides
the most consistent credit scores across the three CRCs. -
There is no statistical bias in VantageScore 4.0-based credit scores
with respect to any protected ethnic groups.
PROMOTING TRANSPARENCY AND MODEL GOVERNANCE
As part of its mission, VantageScore Solutions annually assesses all
VantageScore models and shares the results publicly with market
participants and regulators to promote transparency and support model
governance activities.
The model performance assessment is based on a sample of 45 million
anonymized and randomly selected U.S. consumer credit files obtained
from the databases of each CRC (15 million files from each of the three
CRCs), which were analyzed over a standard two-year timeframe
(2016-2018) using standard statistical tests.
TRENDED CREDIT DATA
The VantageScore 4.0 model has implemented first-to-market innovations,
such as the use of trended credit data, which captures the trajectory of
borrower behaviors over time.
To measure the performance benefits provided by trended credit data,
VantageScore 4.0’s prediction results were compared to those of
previously developed credit scoring models that contain only static
(point in time) credit attributes. Test results show an improvement in
predictive performance as a result of the trended credit data,
particularly in the higher credit score ranges where risk of default can
be more difficult to identify and assess. As a result, users of
VantageScore 4.0 will benefit by their ability to more finely tune
credit offers to these higher scoring, lower risk consumers.
MACHINE LEARNING
VantageScore 4.0 also is the first and only tri-bureau credit scoring
model to leverage machine learning techniques in the development of
scorecards for those who cannot be scored by conventional credit scoring
models (i.e., the Newly Scored population segment).
Within this population of consumers, the performance assessment shows
that VantageScore 4.0 outperforms VantageScore 3.0 in new account
originations (7.4%) and existing accounts (2.3%).
Moreover, the assessment showed no statistically significant difference
in the default rates observed within each VantageScore 4.0 score band
for Mainstream consumers as compared with the Newly Scored consumers.
This finding demonstrates that the VantageScore model, while adhering to
safety and soundness principles, is still able to assign an accurate
score to those who cannot get a credit score when conventional models
are used.
“Despite the healthy economy, it is important to continually assess the
performance of credit score models using a number of parameters to
ensure safety and soundness, as well as fairness. While as a model
developer, we seek to ensure the model is properly rank ordering,
lenders should monitor for absolute risk,” said Barrett Burns, president
& CEO of VantageScore Solutions. “Inevitably, the credit cycle will
change and lenders must be diligent about performing frequent testing
and validation of their existing models to guard against the potential
that risk will increase and cause an unexpected uptick in defaults.”
While we always share the results of our models’ performance assessments
publicly online, we encourage lenders to perform their own model
assessments within their lending environment when considering whether to
upgrade to new models. This year’s assessment can be found at www.vantagescore.com/ModelPerformanceAssessment19.
About VantageScore Solutions
Credit scores can impact many aspects of your life, everything from
whether you are able to get a loan and how much interest you will have
to pay to whether you are able to rent an apartment. At VantageScore, we
understand the impact credit scores have and we take it seriously.
VantageScore Solutions, LLC (www.VantageScore.com)
is the independently managed company that owns the intellectual property
rights to the VantageScore credit scoring models and is the leader in
scoring innovation. Recently introduced VantageScore models score
approximately 40 million more consumers who typically are not scored by
conventional models – without sacrificing predictiveness.
VantageScore credit scores are used by lenders, landlords, utility
companies, telecom companies, and many others to determine
creditworthiness. In fact, a recent study found that more than 10.5
billion VantageScore credit scores were used in June 2017-July 2018. Of
those, approximately 6 billion scores were used by nearly 2,500
financial intuitions and over two billion VantageScore credit scores
were provided directly to consumers through dozens of websites and
lenders who provide their users and customers with their credit scores
for free. By using the VantageScore model, these enterprises have access
to many more consumers, and in turn, consumers have greater access to
mainstream credit.
While there are many credit scoring models in the industry, the
“win-win” for VantageScore is its innovative, highly predictive,
patent-protected, tri-bureau scoring methodology that provides lenders
and consumers with more consistent credit scores across all three
national credit reporting companies.
View source version on businesswire.com: https://www.businesswire.com/news/home/20190606005224/en/