In the past few months, the financial industry has heard a lot about the potential for using machine-learning techniques to analyze risks in consumer lending. Up until now, however, there has been little concrete detail about the specific application of these techniques to credit scoring. Recent developments in credit score modeling have proven that machine learning can be harnessed to significantly improve credit decisioning insights.
It’s true that in the blink of an eye, computers can evaluate vast quantities of data, identify patterns and then pinpoint the most predictive ones. That capability is now a crucial factor as the financial industry works to expand the scoreable population.
How machine learning expands the scoreable universe
Normally, model attributes only incorporate one or two types of behavioral credit data. But machine learning can analyze hundreds of thousands of behavioral data relationships to recognize patterns and discern the most predictive ones. These new insights may then be incorporated into the classic attribute format of a production scorecard.
Now, with a sharper picture of consumer behaviors, the industry can see an individual’s creditworthiness more accurately, even when there is little evidence of a consumer’s behavior in his or her file.
What kind of difference can machine learning make?
Machine learning has considerable potential for consumer lenders: It facilitates the accurate risk assessment of the “credit invisible” population, including consumers with “dormant” credit histories (i.e., consumers who have scoreable trades but do not have an update to their credit file in the last six months). The new patterns arising from the use of machine learning reveal information that can increase the predictive accuracy of the credit score assigned to tens of millions of credit prospects — consumers who in fact may have attractive credit profiles. Credit scoring models with machine learning give lenders a powerful competitive advantage: They can score and target segments of the population that competitors using traditional credit scoring models can’t even see.
Increasing the predictive lift for consumers with dormant credit histories
The first tri-bureau credit scoring model to incorporate machine learning has just come on the market: VantageScore 4.0. VantageScore’s data scientists leveraged machine-learning techniques to develop multidimensional attributes for consumers with dormant credit histories. Importantly, these attributes and scorecards are then aligned and fully integrated into a traditional scoring algorithm with a focus on compliance concerns.
Using machine learning, VantageScore 4.0 drives improved predictive performance (Gini). Gini is a statistical measure of a model’s capacity to identify consumers who are likely to default by assigning low scores to them, while consumers who are likely to remain current over a two-year timeframe receive higher scores. Models with Gini results above 45 are considered to rank order consumers effectively.
The VantageScore 4.0 Gini for consumers with dormant credit histories is 52.1 compared to a VantageScore 3.0 Gini of 49.7, capturing between 2.4 percent and 5.9 percent more bad accounts (respectively) in the bottom 20 percent of the population than VantageScore 3.0 did.
With enhanced accuracy, VantageScore 4.0 scores 30–35 million consumers* who were previously considered unscoreable with conventional credit scoring models. This innovation helps bridge the gap between access to mainstream credit and those consumers without deep credit histories — people who have traditionally been shut out of lenders’ automated underwriting systems.
VantageScore 4.0 is also the first and only patent-protected, tri-bureau credit-scoring model built in anticipation of public record and collection trade suppression associated with the National Consumer Assistance Plan (NCAP) — and the first to use trended credit data from all three credit reporting companies (Equifax, Experian and TransUnion).
As the credit industry looks for new ways to expand the customer universe, machine learning provides a significant competitive advantage and uniquely meets today’s regulatory demands.
* Reduction in public records and collection trade lines in consumers’ files will cause the number of consumers who would be newly scoreable using the VantageScore credit scoring model to decline.