Can advanced analytics for credit scoring change the mortgage market?
Equifax uses trended data to better qualify loan applicants
By Alison Bolen, SAS Insights Editor
For many consumers, fostering a positive credit score can take time, discipline and a lot of financial planning. Paying bills on time, monitoring account balances and minimizing the number of open accounts are just a few of the steps consumers can take on the path to a better financial future.
Consumer credit scores, one measure of creditworthiness, consider a number of factors, including total debt, payment history and the length of credit history. For lenders, these three-digit scores are used as an important factor in determining a borrower’s ability to pay back a loan.
But what if there was a better measure? What if – instead of looking primarily at past behavior – lenders could look at how current behaviors are trending, and predict future behaviors based on those trends?
Equifax, a leading provider of consumer insights, is using advanced analytics to do just that. Developing trended data models that learn from large amounts of data, Equifax has developed a new way to measure creditworthiness that offers two benefits:
- Lenders can safely loan more money.
- Borrowers can have access to credit sooner than before.
“With traditional credit scores, lenders usually don’t consider scores below 620 when using this static attribute,” explains Vickey Chang, Vice President of Global Analytics at Equifax. “Now, with trended data and advanced analytics, we can compare credit files over time to look at this month and the past 24 months. We create a lot of time series attributes and use that to build models that we then compare to credit scores.”
“We strive to be innovative. Machine learning gives us the opportunity to stretch our horizons and provide more value to our customers. We can further differentiate good and bad behaviors and be more predictive using machine learning.”
Vickey Chang, Vice President of Global Analytics, Equifax
Chang leads a team of data scientists at Equifax who have built Equifax Dimensions to differentiate consumers from their traditional credit scores. The technique aggregates raw data like balances, payment amounts, credit limits, balance transfers and spending habits. All of these attributes are combined and compared using time series analysis to understand changes in behavior over time.
The new model has identified that many borrowers with scores below 620 are actually safe to approve based on financial patterns that are trending in a positive direction. “We apply machine learning techniques to build new models, and then run analyses to determine which consumers who were declined with traditional credit scores would have actually been safe for lenders to approve,” says Chang. Her team has applied the advanced modeling techniques to two years’ worth of US mortgage data to determine that billions of declined loans could have been loaned safely. “We can now provide lenders with a more comprehensive consumer picture, which consequently widens the universe of potential clients,” explains Chang.
Outside of mortgage lending, these same techniques can be used for auto loans and other consumer loans. Plus, Chang says machine learning can be used to identify potential customers for marketing purposes. “For example, we can predict consumer likelihood to open any type of account,” says Chang. “We can predict whether the consumer is a high spender or a low spender, and understand their tipping point of spending.”
“We strive to be innovative,” says Chang. “Machine learning gives us the opportunity to stretch our horizons and provide more value to our customers. We can further differentiate good and bad behaviors and be more predictive using machine learning.”
A big data environment for advanced analytics
Equifax has a big data environment that includes Hadoop and in-memory analytics, which make advanced analytics and machine learning capabilities possible.
Machine learning techniques are computationally intensive because they have a tendency to overfit, or find causation among attributes in the data that don’t really exist, so the models need to be tested again and again to overcome this.
“The advantage of machine learning is that it’s predictive,” says Chang. “However, sometimes you can overfit models. In order to course-correct overfitting, we allow different samples, or validate the model under a different time frame.”
The other benefit of machine learning is that it’s fully automated. “When given a new business problem or outcome, machine learning can compete with other analytics capabilities, and we can validate it to see how it works for a specific problem.”
Even with the strong results in mortgage and auto lending, Chang encourages her team to continue to learn and evolve. To help customers make informed decisions, Equifax currently organizes, assimilates and analyzes data on more than 820 million consumers and more than 91 million businesses worldwide.
Additionally, with operations or investments in 24 countries, the team hopes to share US best practices with Equifax colleagues around the world so other regions can repurpose the analytic techniques to work in their own economic, legislative and regulatory environments.
- Brush up on machine learning basics
- Sign up for a free machine learning trial
- Learn how to overcome the challenges of machine learning
Get More SAS Insights
Want more Insights from SAS? Subscribe to our Insights newsletter. Or check back often to get more insights on the topics you care about, including analytics, big data, data management, marketing, risk and fraud.