Credit Risk Policy Management Gets Easier
The single biggest competitive advantage you can develop is the use of effective multiple-data set models. The ability to manage these multiple sets applies equally to making better risk decisions. Specifically, a little bit more accuracy in scale delivers large gains in profiling customers’ risk attributes.
Credit data is no longer all that is needed to determine risk. Credit data is only a shadow of the picture of customers. How are their utility bills paid, are they using a current address? Does this customer have delinquent utilities not yet in collections? Did this person ask for credit from you previously and not stay on top of their bill? You need to see income as well as debt. You need to be able to predict income from the job description. You need employment data as well as credit history. In short, credit history shows only the after-effects of all the other data points.
The knowledge of these data sources and blending them together is where value is really created. In fact the value created can be enormous.
With this augmenting information, your predictive success can improve drops by a few KS points. KS (or Kolomogorov – Smiroff) measurements are a statistician’s yardstick of how well a model predicts behavior and thus risk. In head-to-head analytic comparisons, the KS score is the gold standard in modeling.
A point or two improvement in KS scores can translate to millions of dollars. A single KS point can represent from a half percent to three percent of risky people. A three percent movement on a $1,000,000 household portfolio is going to have an incredible value. Thus finding additional data that move even a couple of percent of customers in the right direction is well worth the mining. Many companies try to find these points with custom models. When custom models fail, it can be a good idea to switch to a multi-data source set of attributes. Let’s see an example.
If a credit file has 10 trades that are 90 days past due, declining a customer is pretty easy to do. If the customer is not past due on either of only two trades, it might be tough to offer them a better rate. If you had additional data confirming they had a nice satellite and cell phone package, however, you might want to see if you can cross-sell a better rate.
It does take a lot of creativity to get to the right attributes from multiple data sources. Being a data company, Equifax gets to know customers really well. Long term relationships with these data sources have let us get really clever.
Recommended For You
Why Proxy Model Explanations of Neural Networks Inevitably Fail For financial service providers, credit risk modeling is all about identifying […]
The Past as a Window to the Future In my last post, we spent time exploring methods to mitigate the […]
May 2020 Market Pulse Series – FAQs Equifax and our expert guest speakers answered dozens of questions from attendees following […]
The economic uncertainty caused by COVID-19 has impacted the automotive industry in complex ways. The entire auto ecosystem has been […]