Just in case you’ve been wondering where all those 70s’ “obscenities” have ended up, you’d better now check in credit analysis rating systems to understand they haven’t gone far.
In the figure above a credit risk model, fitted with a logistic function, has been tested on an out-of-sample credit file of 500 firms. With obviously good results indeed (see the ROC curve approaching the top-left corner), except for the fact that the out-of-time component of the test had been completely left out; consequences: sensitivity and specificity plummet the following year, which it happens to be 2012. It’s been a long time since ROC and roll.
Leave a Reply