There was an article at BP recently that showed xStats are about as predictive as stats like FIP. The article set out to show that the xStats weren't very useful, but it was pretty flawed. For one, they actually were a little more predictive on the whole. Secondly, the whole idea of xStats is to strip away context like defense and park effects. So, basically, Jon Gray's xwOBA isn't going to be good at predicting his ERA next year if he is still pitching in Coors, and the same with Jon Lester pitching in front of the Cubs' defense. Also, xwOBA correlated much better with same-year wOBA than other stats correlate to same-year stats. So, xStats are just as predictive as something like FIP, but also do a better job of explaining things that have already happened. They seem pretty useful to me.