Why old models are just bad for business
Every few years the companies that design and develop credit scoring models will go through the lengthy process of redevelopment. For comparison purposes, think of how Apple or Microsoft will periodically release a new operating system for your Mac, PC or smartphone. The newer version is very similar to the previous version, with some notable improvements.
As with smartphones, banks and other users of credit scores have the option to convert to the newest scoring model once it becomes commercially available. The process can be time-consuming, given the amount of analysis that must be done to properly adjust underwriting standards for the newer version, as well as the operational changes and governance steps involved. But the benefits far outweigh the challenges.
Newer credit scoring models are more effective than older scoring models and that’s really their primary function: predicting a future default. Changes in consumer behaviors, lending practices and the macro environment are better represented in newer models that are built on more recent observations. Further, the availability of new forms of data and more advanced modeling methodologies provide additional firepower to a newly developed model.
There are a variety of empirical methods used to compare the performance of credit scoring models. The Gini coefficient (“Gini”) and the Kolmogorov-Smirnov (KS) statistic are two common metrics used to compare the effectiveness of credit scoring models and their ability to delineate between future “good” and “bad” credit risk populations. Lenders perform such “champion-challenger” comparisons in evaluating new models. Other analyses performed to assess the performance of new credit scoring models relative to an older model compare profiles of approved and rejected borrowers, resulting in default rates as well as approval rates have given a loss tolerance.
Failing to adopt the newer versions means lenders are making decisions based on less effective tools, leading to suboptimal risk and pricing decisions. And, choosing to continue to use much older scoring systems, like what is happening in mortgage lending because of FHFA policy, is counter to sound underwriting and risk assessment practices.
Captures More Recent Trends in Risk Assessment
A valuable product of newer scoring systems is the ability for the model developers to react to more recent data and research and development findings, and then apply these learning into their scoring models. This is commonly referred to by credit score experts as “following the data.” The underlying premise of a credit scoring model, like any model, is that what is observed in data historically will be indicative of the future. Hence, models that take into account more recent observations are more representative of the “through the door” populations and will perform better. Newer models also benefit from insights gained from data elements that were previously unavailable for modeling.
There are many examples of features included in newer scoring models that are not captured in older scoring models, like the ones used by mortgage lenders. Some of these features in newer models relate to:
- Changes in borrower behavior post-Great Recession
- Treatment of smaller dollar collections or collections with zero balance
- Treatment of medical collections
- Accounting for the reduced presence of derogatory public record information due to changing reporting requirements, and
- Use of new data measuring changes in credit behavior over time versus a static snapshot
Benefits Consumers and Lenders Alike
As previously addressed, newer scoring systems do a better job of more clearly separating lower risk consumers from higher risk consumers. What this means, in practical terms, is low-risk consumers are going to have higher credit scores under the newer scoring models. Think about that: Higher credit scores simply because a lender is using a newer and better scoring model. This translates to higher approval odds for a loan and better product terms, such as lower interest rates and higher credit limits for creditworthy borrowers.
The opposite, of course, is true as well. In older scoring systems, low-risk consumers are going to score lower than they deserve. This means consumers who do business with lenders that use older scoring models may have lower chances of being approved for a loan or face less favorable product terms, compared to what they would receive from lenders that use newer, more effective scoring models.
For lenders, the choice is between growing their business safely and soundly with the help of a newer and more effective scoring model supporting their underwriting decisions, or using an older model that may lead to loss of good business to competition, or worse, originating loans with higher risk of default than desired.
At the end of the day, lenders and consumers both win from using newer credit scores. Matching creditworthy consumers with the right loan products is good business for lenders and helps consumers meet their needs while guarding them against getting in over their heads with debt.
No Reason Not to Use Newer Models
There really is no good reason not to convert to the newer scoring systems if this helps business.
Sure, there’s work involved in converting to a new credit score. This may include an analysis that needs to be performed to determine the adjustments necessary to underwriting policies and product features, going through required governance processes, such as credit committee approvals and model validation, operations and technology changes, training and communication for staff, among others.
These activities are not uncommon as part of the normal course of business and should not deter from implementing a newer and better model. Lenders go through similar steps when introducing a new product, adjusting policies, entering new markets or adopting new processes or technology. While there is an initial investment, longer-term benefits offered by a newer model far outgrow the cost.
Disclaimer: The views and opinions expressed in this article are those of the author, John Ulzheimer, and do not necessarily reflect the official policy or position of VantageScore Solutions, LLC.