How to Pick a Credit Model

Subscribe to Our Blog

Mobile check deposits, robo-advisors, video tellers. So many innovations profound and mundane have swept through the financial services industry in the last decade. One arena that remains relatively unswept, however, is credit risk modeling. The science of predicting good borrowers from bad still leans heavily on the FICO score, a decades-old logistic regression technique that combines a few dozen variables in clever but fairly simple math from the 1950s.

Data Screen Machine Learning

Banks and other lenders are increasingly looking beyond FICO to machine learning (ML) to predict credit risk. Machine learning allows lenders to predict risk more accurately by harnessing lots of data and advanced math that takes advantage of today’s nearly unlimited compute power.

For lenders, the billion-dollar question is: What’s the right modeling approach for my business? Logistic regression is safe and explainable. Machine learning promises better performance: more approvals with no added risk or same approvals with reduced risk. Not all machine learning techniques are the same, though. Some work well in the testing phase, but cannot be deployed for everyday use because no one can explain what’s going on inside the model sufficiently enough to satisfy regulators and compliance folks.

Some approaches try to split the difference by enhancing logistic regression with a smidgen of machine learning to get slightly better predictive performance. Other modeling packages tout their ability to automate the data science part of the problem, but leave you with shelfware models that, again, you can’t put into production because they lack the necessary explainability and model risk management workflow. ZestFinance’s ZAMLTM underwriting solution offers comprehensive model risk management tools, including automated documentation, testing, and production monitoring required to meet all regulatory requirements.

Before you commit, it’s crucial to know about each technique's financial performance, explanatory power, and operational advantages. What kind of risk reduction can you expect? How accurately can you assess its impact on protected classes? Is it easy to make adjustments as market conditions change?

Whether you’re buying or building a new ML credit model, this scorecard we put together can help you navigate the options and how they perform against key requirements.

Machine Learning Scorecard ZAML

Click to enlarge the scorecard thumbnail. Or click the button below, and we’ll send you the entire scorecard booklet. Questions? You can always reach us at partner@zestfinance.com.

 Receive the Scorecard

 

If you enjoyed that, you may also enjoy our article on Explainable Machine Learning in Credit