Hey ben, I have a pseudo-related question which is: how do you do your forecasting?
The way I do it is that I get all the minute adjusted numbers for usage, reb%, to%, ast%, def% and ts% and also the sum 3PM and fls/g. Then I calculate the rank for each team against the other teams in those categories (except for usage). There is a final ranking for each team which is .25(ts%)+.25(def)+.5(avgrank for 3PM, reb%, ast%, to%, fls/g). From there, I apply there resulting rank to the average wins for the person who finishes in that spot over the number of seasons that I track.
The last thing I do then is tweak them based on how I feel about the coaches and if they are outside of some reasonable minutes or usage tolerance when I do the over/unders and other things I write up..
It isn't perfect, especially the qualitative adjustments I make at the end which I should probably get rid of. But the raw ranking output from last season was really spot on in terms of what order people finished in last season. (I deleted the info to fill in this season's but I think it got the top five and bottom three correct in order iirc) I think the flaw in this design though is when people operate on the fringes and go to the extreme in a particular area (example being your current team in rebounding for this season). I don't really know how I might account for that sort of behavior in any sort of quantitative forecasting.