During the six days our family was without power after Hurricane (or is it Superstorm?) Sandy swept through New Jersey, my wife worked very hard to keep our kids happy and busy. Before the storm hit, she went out and bought some new board games for us to play, in case we were without power for an extended period. One of the games she bought was Mouse Trap. The game involves building a Rube Goldberg trap to catch game pieces shaped like mice. My wife spent 20 or 30 minutes building the mouse trap before declaring we were ready to start playing. I pulled out the instructions to figure out how the game was played, and then realized that the mouse trap is built as the game is played, not before. So after laughing for a few minutes, we took the game apart and started over again. (For those of you keeping score at home, I won.)
Building a better mouse trap without using or understanding all the information that is made available can be very difficult. The same is true in loan underwriting. Data is important, but so is context. A former chief information officer at Google is trying to incorporate machine learning and predictive modeling into a loan underwriting model to better assess a consumer’s credit risk.
The new company is called ZestFinance, and the underwriting program called Hilbert (named after statistician David Hilbert). The company is currently working in the payday lending industry, but it’s easy to see that the model could be applied to any consumer (and even commercial, perhaps) lending sector, including auto finance.
The credit bureaus have been so reluctant to share information about their formulas and models that it’s difficult to compare them to what ZestFinance says it is doing. What ZestFinance says is that its model is giving more loans to more people while improving default rates. As proof of concept, the company has raised nearly $73 million in funding to date.
There’s always been a contextual component to subprime lending. One banker once described subprime lending as “story loans,” because every subprime borrower had a story behind why his or her credit scores were low. Medical problems, family problems — there was usually a reason. Assessing the human component of credit risk has always been the hardest part to loan underwriting. ZestFinance claims it can do that. Whether true or not, it could signal a new evolution in consumer lending. One in which everyone may need a new set of instructions.