2012-2013 predictions

[mystic]

Right now I try to get a good system which can predict minutes based on various variables. Once I have that, I can give a prediction for this season. Given the fact that this would be entirely based on data until last season, it should be a problem at all.

[v-zero]

Right now I try to get a good system which can predict minutes based on various variables. Once I have that, I can give a prediction for this season. Given the fact that this would be entirely based on data until last season, it should be a problem at all.

It would seem to me that a neural network is the ideal framework to predict how long a player will play for in a given season.

[Crow]

http://basketballprospectus.com/article ... cleid=2433 for the Schoene projections.

This season's predictions from the guy with the best projections published on the net last year according to WILQ
http://high5.rivals.com/content.asp?CID=1176796
(I was second in that broader comparison.)

Is there a list of Bill Simmons' win projections?

[Crow]

Not exactly sure where I'd put Minnesota but some of the projections for them seem high to me.

Schaller's picks generally seemed pretty good. One exception: 44 for Minny seems at least a bit high.

[v-zero]

Rubio, Kirilenko, Love... Those three are worth 20+ wins on their own, including injuries.

[Neil Paine]

I added Schaller to the table on p1.

[DSMok1]

Could you update with my revised projections from the 30th, Neil? I fixed some issues with the ESPN fantasy projections for playing time.

[Neil Paine]

Could you update with my revised projections from the 30th, Neil? I fixed some issues with the ESPN fantasy projections for playing time.

Added.

[Crow]

Thanks for compiling Neil.



MIN, OKC, ORL and UTH have large ranges in the predictions. 10+ wins. Any others that I missed?

Though Baskebetball Propectus' final projections cut the Timberwolves title chances from what I saw in ESPN the magazine (almost 10% title chance) I am still surprised to see more than a 5% chance given there for them. I wouldn't give them more than 1% chance.

[DSMok1]

Correlation Matrix for the Predictions.

The Las Vegas and ESPN predictions are very similar. Wages of Wins are very different from anyone else. Schoene is pretty unique; JH is quite similar to TPR (What is TPR?). My ASPM predictions are most similar to the "dumb" predictions (similar methodology, different stat) and Team Rankings.

[J.E.]

My current power rankings based on minutes played in all games so far, and RAPM, agree the most with "dumb" and WoW2. (Taking the absolute difference between expected_PD*2.5 and # of wins given by each projection). It disagrees the most with JH. Biggest single disagreement with one team projection is with WoW2 and Minnesota (52.2 vs 26.8)

It has to be noted that currently injured players (everyone listed as "questionable" or worse at donbest) were projected to play 0 minutes for the rest of the season. Players that haven't played yet were also projected to play 0 minutes.

[Crow]

Bobbofitos commentary:

http://pointsperpossession.com/nba/34-n ... rt-eckstut
http://pointsperpossession.com/nba/34-n ... rt-eckstut

Magic off to a surprisingly good start.

Bargnani and Terry off to slow starts.

Hornets, not so bad.

Mo Williams better than I expected and I agree Utah should do pretty well.

Spurs defy efforts to write them out.

[Neil Paine]

Checking in on these after a couple weeks:

Code: Select all

----------------------------
Projection      Mean Abs Err
----------------------------
"dumb"          5.4
TmRkgs          5.5
DSM_1           5.6
DSM_2           5.7
WoW2            6.2
WoW1            6.4
Bobbo           6.7
v-zero          6.7
ESPN S.F.       6.8
SCHOENE         7.1
Bovada          7.1
LVH             7.3
TPR             7.3
JH              7.9
----------------------------

Here's how I'm measuring the accuracy -- I'm calculating each projection's average mean error vs our best Bayesian guess at what each team's final win total will be, assuming a prior mean wpct of .500 and stdev of .145. You can get the team's expected record the rest of the season by taking the team's actual record & adding 12 games of .500 ball. So if the Knicks are 5-0, we'd expect them to play (5+6)/(5+12)=.647 ball the rest of the way, so their expected end of season win total would be 5 + .647*(82-5) = 54.8. Do that for every team, and that's what I'm measuring everyone's projections against.

[DSMok1]

Should probably post RMS error, Neil. That's how predictions are usually tested, aren't they?

[J.E.]

Current power rankings, with injuries (Wade out is the reason for MIA so low)

Code: Select all

NYK	7.9851405569
SAS	6.8094876755
LAC	6.5730003985
MEM	6.444796111
LAL	6.4153382673
DEN	5.2736314728
OKC	4.255923397
MIA	4.1095466209
CHI	3.1028168956
ATL	2.1961780408
MIL	0.9920235733
HOU	0.9824112452
BOS	0.7909110357
UTA	-0.4361178615
BRK	-1.1125038609
IND	-1.8520208862
DAL	-2.702598092
MIN	-2.83493091
PHI	-3.047900583
GSW	-3.1168699687
NOH	-3.1495879543
CLE	-3.3740691139
POR	-3.550685337
SAC	-4.4660374671
PHO	-5.0499664605
TOR	-5.2706958786
DET	-5.3445856337
WAS	-6.8914262492
CHA	-7.2418014073
ORL	-7.7446777043