Insider article of his from around 4 years ago.
He found that Summer League is completely worthless for 2nd year players, but surprisingly useful for rookies:
Veteran players who were in the league the previous season have an even smaller correlation of .101 between 2012 summer performance and 2012-13. Once we account for how well these players were projected to play in 2012-13, their summer-league stats have zero predictive value.
By contrast, the correlation of .463 for rookies is far higher. In fact, it's nearly as strong as the relationship between my college stat translations and rookie performance (.468).
The idea that a Summer League game matters way more than an NCAA game actually makes intuitive sense - you're playing by NBA rules (3 point line, shot clock, etc), against borderline NBA talent. In the NCAA, you are playing against total scrubs (by NBA standards) for an overwhelming majority of the time.
It is surprising though that ~5 Summer League games are around as predictive as 30+ NCAA games. I wouldn't have expected it to be that drastic.
