Crunching data from Prosper, the peer to peer lending site, researchers correlated default rates with words most likely to appear in a loan application. From an article in New York magazine:
Somewhat intuitive when you look at the types of words (the former suggest a degree of financial planning; the latter hopes and prayers) but still fascinating. And what about using data like this to reverse-engineer “better” loan applications? Of course at scale that would start to skew the data overall, but done selectively you could create “false positive” loan applications by mimicking the language from “good” loans. Sorta how I’ve always wondered if successfully-funded Kickstarter projects have project descriptions that are substantially different than those which fail.