APBRmetrics

Now that the regular season is over, I want to propose that each of us with rating systems do a retrodiction of 2010-11 win totals using 2009 player ratings and actual minutes played in 2010-11. Actually, you could use any prior-year rating system you like. The goal would be to see who comes up with the most accurate retrodiction results. Obviously, the only criterion is that you can't use any 2010 data besides minutes.

I'd like to see ASPM, eWins, RAPM, etc, in addition to mine.

Are you guys up for it? I think we can learn a lot.

How do you want to measure error? I would prefer squared difference in actual points scored vs forecasted points scored per possession

Sure, I'm up for it. I already had preseason predictions for ASPM tabulated, so it's trivial to adjust them with actual minutes played.

I'd like to use regular season only (obviously).

For the primary measure, I'd use team efficiency differential, adjusted for opponent. I've got those numbers calculated already.

Yeah, I was thinking squared error in differential, too.

Regular season only.

Also, think of this as an iterative challenge. You can submit as many results as you like. The goal is to get better and better, if possible...

Or for someone like me, maybe just to see how far behind I am.

Also, we should use a "master" possession file so that everyone is working with exactly the same data.

I don't really like team efficiency differential. One might predict a good differential for Boston because his system says Glen Davis is great. Per possession error is more useful I think. I'll go along with team efficiency differential though

http://www.basketballvalue.com/publicda ... 110416.zip
should be the masterfile

EvanZ wrote:... do a retrodiction of 2010-11 win totals using 2009 player ratings and actual minutes played in 2010-11... ... Obviously, the only criterion is that you can't use any 2010 data besides minutes.
.

Clarification needed. To me, 2010 refers to the 2009-10 season. That's what I'd always use to predict the 2010-11 season.
If by 2009 you mean 2008-09, to predict 2 years later, that's kind of strange.

J.E. wrote:I don't really like team efficiency differential. One might predict a good differential for Boston because his system says Glen Davis is great. Per possession error is more useful I think. I'll go along with team efficiency differential though

http://www.basketballvalue.com/publicda ... 110416.zip
should be the masterfile

Oh, you mean based on the actual lineups? That would be tougher to run, but perhaps more accurate. I could do that.

J.E. wrote:I don't really like team efficiency differential. One might predict a good differential for Boston because his system says Glen Davis is great. Per possession error is more useful I think. I'll go along with team efficiency differential though

http://www.basketballvalue.com/publicda ... 110416.zip
should be the masterfile

Well, assuming it would be trivial for you to do team differential as well, can we do both? I might just participate in the team differential because it's less work, but maybe would try the lineup level too at some point.

Would that be ok with everyone to have both?

Mike G wrote:

EvanZ wrote:... do a retrodiction of 2010-11 win totals using 2009 player ratings and actual minutes played in 2010-11... ... Obviously, the only criterion is that you can't use any 2010 data besides minutes.
.

Clarification needed. To me, 2010 refers to the 2009-10 season. That's what I'd always use to predict the 2010-11 season.
If by 2009 you mean 2008-09, to predict 2 years later, that's kind of strange.

2009-2010.

In the past I've had to explain to, for example, WP'ers that using current season data to "predict" current season wins isn't a prediction. I probably didn't need to make that explicit here given the crowd.

I'm game! I'll just use my statistical prediction of RAPM...

I bet Daniel's wins since his inherently predicts future performance.

bbstats wrote:I'm game! I'll just use my statistical prediction of RAPM...

I bet Daniel's wins since his inherently predicts future performance.

ASPM was not tuned for OOS. However, I probably have a more mature projection framework than most, yes.

Are we allowed to also use '09-'10 playoff data in our predictions for '10-'11 regular season?

I think we should only do per-possession predictions, due to trades & team strength of schedule.
But that means I probably need to regress my ratings for players with few possessions played...don't know how to do that just yet. D'oh!

Sure, use playoff data. Anything known before opening night is fair game.

EvanZ wrote:Now that the regular season is over, I want to propose that each of us with rating systems do a retrodiction of 2010-11 win totals using 2009 player ratings and actual minutes played in 2010-11. Actually, you could use any prior-year rating system you like. The goal would be to see who comes up with the most accurate retrodiction results. Obviously, the only criterion is that you can't use any 2010 data besides minutes.

I'd like to see ASPM, eWins, RAPM, etc, in addition to mine.

Are you guys up for it? I think we can learn a lot.

How do we deal with rookies? Do we use actual rating results (in 10-11) for them?