Did over-reliance on models and formulas cause bankers to shut off parts of their brains, making them miss some obvious risks?
Mike at Rortybomb offers up some pretty compelling slides demonstrating how and why the answer is ‘yes’. The example here is the straightforward FICO score, the personal credit rating that everyone is familiar with. Turns out that the GSEs (Fannie and Freddy) established a score of 620 as cutoff for reselling certain loans into a CDO. Above 620 and you could sell of the loan, no problem. Below 620 and you had to carry more risk. The upshot is that if you were going to make a loan to someone with a FICO score in the 615-619 range, you had to do more due diligence than simply looking at their FICO. You had to employ more “soft skills” or even intuition to figure out loan-worthiness.
What the chart below shows, rather clearly, is that defaults are actually higher in the band above 620 than below. A banker forced to use judgment rather than rely on a single numerical score will make fewer bad loans, even if it means they’re making loans to someone that a model has deemed less credit-worthy.
(For what it’s worth: There is an alternative theory, which is that people who cluster right above this threshold may have intentionally gamed their score in the months leading up to the loan application, suggesting that the score itself is more reflective of recent FICO-focused activity than the person’s overall creditworthiness.)