Tom Jacobs

A Better Corporate Credit Risk Model Still Not in Sight

Unfortunately, mathematical models and limited data are typically all we have save for those who are practicing loan officers. FICO scores work well for many problems. It is the assumptions around correlation between borrowers that are most at fault as well as the complexity in combining multiple instruments separating the ultimate lenders from the borrowers. While subprime may have gone too far and received far too much government backing, I still don't believe we want to go back to telling the lower middle class they can borrow at pawn shops or not at all.

While I won't dispute a number of the criticisms the author has provided for building a good model or the failure of most practitioners to understand, collect, clean and appropriately use their data with a given model, something needs to be done differently and anything open source is at least a step in the right direction. Some credit should be given for at least trying to change from the status quo.