All this week, we’ve been looking at the data and algorithms behind credit scores. While many lenders will use FICO scores, the company has a major competitor, VantageScore.
It was founded by the three credit bureaus – Experian, Equifax and TransUnion – in 2006. The independently managed company says its scoring model is more inclusive and predictive of credit risk than traditional models.
Marketplace Tech host Kimberly Adams recently spoke with Silvio Tavares, President and CEO of VantageScore, about what he and his team consider when designing their algorithms. The following is an edited transcript of their conversation.
Silvio Tavares: The reality is that establishing a person’s creditworthiness requires a great deal of judgment. It’s really important [to consider] decisions about what data to consider – and how. Let me give you an example. VantageScore was one of the first companies to come up with the idea of eliminating paid medical debt. Two-thirds of collection accounts were related to medical debts. And when we looked at the data, we found that if you had indeed paid off your medical debt, the mere fact that you had a medical debt collection really didn’t indicate anything about creditworthiness. We also determined that many of these paid medical debts disproportionately affect people of color. These are the main types of decisions that [as] modelers, we must take into account. It’s about getting the best objective data and empirically researching what kinds of data sets are best for empirically predicting whether someone is creditworthy or not.
Kimberly Adams: What do you do at VantageScore to ensure that all data that powers your algorithm and model comes from consumers who have knowingly agreed that you may have this information?
Tavares: We work with banks and lenders, as well as fintech lenders, to ensure that they collect data transparently and that the consumer has provided consent. But actually most of that will be done by the bank, it won’t be done by VantageScore. We focus on the algorithm and ensure that it is fair, equitable, and most importantly, it is predictive, and is a very good predictor of consumer reimbursement.
Adam: How can consumers be informed of what this consent really means?
Tavares: We’re not looking specifically, OK, what types of stores did you shop at? You know, did you buy big wheels for your car from a reputable dealer? We’re not specifically looking at that. We look at broad metrics of your overall financial health and cash flow. But it’s also fair to say that the reality is that the market is moving very quickly. The story of credit models is that they are not based on this bank account type data. But probably over the next 10 to 15 years, bank account data will be one of the primary sources of credit reporting, simply because it’s timely, often more accurate, and gives a much better picture of actual consumer behavior. Many of the consumers who will benefit the most are the new type of consumers, the consumers who have been historically underserved, who are going to be able to access better credit terms, access more credit products, thanks to these new algorithms that we develop .
Adam: What role does machine learning play in all of this?
Tavares: Machine learning is a general term, which is a subset of artificial intelligence. And that basically refers to looking at data – usually unstructured data – and then using that data to come up with a prediction. You know, machine learning is a very controversial topic. And that’s the one we really focus on a lot. But we’re really focused on analytics and using advanced analytics, all kinds of different analytics techniques, including machine learning. And we’re looking at ways to improve the predictive power of our scores. But also, importantly, include more people. That’s why we pioneered and included things like rental data, which isn’t a loan obligation, but it’s a really good predictor of what you’re going to do. So we think algorithms are about making smart choices about how you use data to include more people, but also, very importantly, understanding how that data will effectively predict consumer behavior.
Adam: Your conversation about rental data reminded me of a question I wanted to ask you earlier, as you try to include this rental data, it can be confusing as some landlords don’t provide data on who rents or pays rent on time, consistently, especially when it comes to some of these more vulnerable groups like recent immigrants or low income people. What have you all done at VantageScore to try to get access to this kind of consumer data for the most vulnerable people, who may be paying their rent in cash to a landlord who may not be not be the most encouraged to report it?
Tavares: The first thing we did was we were the first to integrate this rental data when it is provided to the credit bureaus, right? The second key thing we do is look at other ways we could potentially get this data. Many consumers, especially consumers with limited incomes, are paid on payroll cards, which are a type of credit card or debit card. We are looking at how can we access data from these providers and incorporate it into our algorithm. We also review transactions on bank accounts and debit accounts. And if the consumer gives us permission to access them, we can also use them in our algorithms. The reality is that there are so many different types of data sources today, but there are more datasets than we have time to review. And so it’s really about prioritizing those so that we can incorporate them and get the best datasets with consumer permission. This will open up this really true view of solvency.
You can read more about how VantageScore designed its latest credit score model, 4.0, around the concept of financial inclusion here.
VantageScore and FICO are considered the two main competitors in the credit scoring market. This Credit Karma explainer lays out some of the key differences between the two.
But we should also note that FICO also updates its models. The latest, FICO Score XD and UltraFICO, “responsibly leverage alternative data sources to help lenders identify new credit-ready consumers and increase access to credit not just in the United States but around the world.” world”.
And we also want to point out that lenders can select which scoring models they want to use. The ones we may see as consumers are not always the scores used to determine if you get a loan.
Comments are closed.