News

The Difference A Credit Scoring Model Can Make

Credit scores are a crucial piece of the lending pie. But scoring models that came of age in the 1980s and ’90s don’t work with a 2015 post-recessionary consumer. Sarah Davies, VantageScore SVP sat down with MPD CEO Karen Webster to talk about what lenders are leaving on the table, and what they can pick up without incurring any more risk – just by rethinking how they score credit risk.

 

Credit scores are a crucial piece of the lending pie. But scoring models that came of age in the 1980s and ’90s don’t work with a 2015 post-recessionary consumer. Sarah Davies, VantageScore SVP, sat down with MPD CEO Karen Webster to talk about what lenders are leaving on the table, and what they can pick up without incurring any more risk – just by rethinking how they score credit risk.


KW: What is different about the post-recessionary consumer and why don’t existing credit scoring models work?

SD: The reality is that all of these models are working. Credit scoring models provide rank orders, meaning customers who have a higher likelihood to pay their debts are placed at the top of the credit scoring range, with scores of 750 and above, while those who have a lower likelihood of paying their debt fall at the bottom of the range — this is what is known as rank ordering. Almost all of the credit score models available in the industry today are rank ordering, but the question then becomes, “how well do they rank order?”

A perfect credit scoring model will put all of the people who pay down their debts at a score of 850 and all of the people who do not at a score of 300, but because we live in a world of variability and statistics, credit scoring models are actually unable to rank order perfectly. In an environment that is post-recessionary, credit scoring models should be using the most up-to-date information to rate consumers to ensure they are being ranked as perfectly as possible. The issue is that models many lenders are using were built using the credit behaviors, consumer behaviors, and banking products that were relevant in the early 2000s, which are not going to provide effective rank orders.

Ultimately, outdated credit scoring models are not doing an effective job because they are not as sensitive to consumers in this post-recessionary window. For example, we have all come through a very painful time in homeownership; hopefully we are on the upside of all of that, but still there is a resistance to taking on large mortgages, high interest rates, and longer-term loans that was not present 10 years ago, resulting in slightly different behaviors. The fact that an older model does not capture these new behaviors and perspectives on mortgages means the model is unable to be as accurate in the way it rank orders consumers. Therefore, in order to get the best performance for the industry, credit scoring models built on relevant time frames such as post-recessionary data, must be used.


KW: Is the data you capture any different than the data the more traditional credit scoring models are using?

SD: The VantageScore model sits on the credit bureau credit file data, which means we use all the data traditional credit score models use. Where VantageScore differs is that if there is additional data available in the credit bureau files, such as data associated with telecommunications or utility information, we will utilize that data as well. Our models also incorporate rent data.

In general, very similar data sets are used in our credit scoring model as other traditional models, but if there is incremental information out there in the credit file we incorporate that as well.


KW: What are some of the outcomes of using different data? The incremental information is available to everyone, but many are not using it while you are. What are some of the differences in terms of outcome as a result of this additional data?

SD: One of the more obvious results of using incremental data is being able to provide a credit score in some situations where it would otherwise not be available. There is a small population of U.S. consumers who do not have traditional data on their credit files, meaning they do not have credit cards, vehicle loans or home loans, mortgages, etc. But these consumers may be paying rent on an apartment, have a cellphone bill or be paying a utility bill. If that is all there is available associated with how they are paying their debt and it is reported to the credit bureaus, VantageScore is going to pick up that data and provide consumers with credit scores. Under any other credit scoring model, these consumers would not be scored, and not having a credit score makes it very difficult to gain access to mainstream lending. A lack of a credit score also has the potential to place a consumer at risk for exposure to predatory lending or pay-day type lending because conventional banking systems will not have a statement of risk on the consumer, which is what the credit score stands as.

Another immediate and beneficial outcome of using incremental data is that when it is added onto conventional or traditional credit file information, it provides lenders with a better and broader picture of the consumer. The information can support a more comprehensive understanding of a consumer’s debt picture and how they are repaying those debts. For example, if a consumer has telecommunications or utility data on the credit file and they are paying those bills on time, they should receive credit for that and their score would improve as a result.


KW: Are you seeing interest on the part of particular types of lenders for using this model? Where exactly is the interest coming from?

SD: The interest is coming from across the board actually. Coming out of a period where the industry as a whole was relatively shut down in terms of giving access to credit, it is interesting to see within the last several years every bank I have talked with is asking how to open up credit again and give greater access to credit. VantageScore 3.0 is able to not only score the traditional or conventional credit-eligible population in the U.S., but also score an additional 32 million to 35 million consumers. With banks seeking ways to gauge the risk assessment on a larger population in order to open up their businesses, there is a greater interest in the VantageScore 3.0 model.


KW: What is involved if lenders who are using the traditional model want to adopt VantageScore 3.0? Is that a painful process? What has to happen?

SD: It is actually not a painful process, though it seems like it would be because in many ways the generic scores that have been used in the past are deeply embedded in banks’ strategies. But what a lender has to realize, and most of them obviously do, is that the score is simply a proxy for a statement of risk.

When implementing a new credit score model, lenders have to recognize how the consumers they work with will be scored under the new model, whether they are going from one brand to another, such as credit bureau score to VantageScore 3.0, or moving within brands, like VantageScore 2.0 to VantageScore 3.0. This means determining the distribution of consumers into each score band, the population distribution of those scored and the default rate distribution. It is important to know how many people fall within a score band, such as how many consumers receive a score between 600 and 620 versus 621 and 640, as well as how many of those consumers defaulted. Being aware of this information as it relates to strategy is what is necessary for the lender to consider when changing credit score models. This is where VantageScore wants to pull back the sense of confusion and complexity associated with this process.

When considering a new scoring model, lenders will also want to understand their score cutoff, which is the lowest credit score a consumer can have in order for the lender to originate a loan, and how it relates to a particular level of risk the business can tolerate. For example, is a lender moves its score cutoff to 700 under a new model, they must determine if they will have the same level of risk, a better level of risk, or a worse level of risk. Ideally, moving to a new credit score model will lower the lender’s level of risk at that particular score cutoff for its strategy and it will enable more people to be approved as a result. The essence of the process to convert from one model to another is simply the lender trying to relate its score cutoff to the default rate and the volume of consumers it plans to approve. As lenders work through the implementation, addressing these factors becomes a frequently repeated exercise, and at the end of the day a relatively simple one.


KW: Are you finding institutions that have implemented VantageScore are actually seeing the outcomes described? Have they been able to increase the throughput of borrowers without assuming more risk and impacting default rates?

SD: Yes, I have heard some anecdotal information, but lender strategies are highly confidential, as it is their way of beating the competition. What we have come to understand is some have the same level of throughput and originate the same volume of consumers but at a lower level of risk under our model. They are still able to bring on all of those consumers but have lower losses associated with them, which is a win for them. Then there are some who were able to keep the same level of risk, while actually originating more consumers and maintaining a larger throughput. We are seeing and hearing examples of both cases.


KW: One of the things you mention in your new white paper is something about the Plug & Play method. What is that?

SD: The Plug & Play method is intended to fight the notion that implementing a new credit score model is highly complex and fraught with challenges. When looking at implementing a new model, if the population distribution between the old model and the new model is fairly similar and the organization is able to see a lower default rate, VantageScore determined the quality of the strategy itself is not being changed in any significant way.

In those situations, credit issuers can just plug into the new credit scoring model and play straight away. There is no major complexity or a lot of risk exposure as a result of it, the idea behind Plug & Play was if there are very similar population distributions, a similar default rate profile, and perhaps a little better score cutoff, then the new model can get up and running right away.


Sarah Davies

Sarah Davies

SVP Product Mgmt & Analytics at VantageScore Solutions

Sarah Davies is senior vice president of analytics, research and product management at VantageScore Solutions LLC, a company owned by the three national credit reporting companies (Equifax, Experian and TransUnion) to provide credit grantors a highly predictive, universal credit scoring model. As the leader of the team responsible for developing VantageScore Solutions’ credit scoring models, Davies and her team are credited with a number of important industry breakthroughs such as a “scorecard” designed specifically to deliver a highly accurate credit score to those with sparse credit histories, and the elimination of paid third-party collection accounts.

In addition to leading the product strategy, she engages with numerous government regulatory agencies and the secondary market to provide insights on consumer financial trends, credit and risk analytics. She is recognized as a national expert on credit scoring technologies, model governance, decision-analytics and is a frequent speaker to the financial service industry and the media on these topics.

She has nearly 25 years of leadership experience in strategy, analytics and information sciences in the financial service industry. Prior to joining VantageScore Solutions, she served as Chief Analytic Officer for iQor Holdings Incorporate, a global customer service and receivables management outsourcing business.

——————————–

Latest Insights:

Our data and analytics team has developed a number of creative methodologies and frameworks that measure and benchmark the innovation that’s reshaping the payments and commerce ecosystem. In the December 2019 Mobile Card App Adoption Study, PYMNTS surveyed 2,000 U.S. consumers for a reveal of the four most compelling features apps must have to engage users and drive greater adoption.

Click to comment

TRENDING RIGHT NOW