2016-17 Team Win Projection Contest / Discussion

[Mike G]

Back to 3 co-leaders.

Code: Select all

3/03   RMSE      avg abs err             SMRE
trzu   4.27      Crow   3.41      kmed   2.68
BaDo   4.39      trzu   3.43      Crow   2.80
lnqi   4.51      BaDo   3.46      yoop   2.91
Crow   4.51      kmed   3.52      trzu   2.92
vegas  4.63      yoop   3.64      BaDo   2.93
RyRi   4.64      lnqi   3.65      lnqi   3.19
cali   4.71      vegas  3.73      vegas  3.22
kmed   4.73      cali   3.77      Mike   3.22
sbs.   4.75      RyRi   3.80      shad   3.22
amp5   4.77      sbs.   3.84      cali   3.24
yoop   4.81      amp5   3.87      ncsD   3.27
ATCt   4.86      ncsD   3.89      RyRi   3.27
sndi   4.91      shad   3.91      jg34   3.29
ncsD   4.91      Mike   3.91      sbs.   3.31
shad   4.93      sndi   3.99      amp5   3.37
Mike   5.06      jg34   4.04      sndi   3.48
AnJo   5.09      ATCt   4.24      ncsB   3.73
jg34   5.28      ncsB   4.35      ATCt   3.86
taco   5.29      AnJo   4.41      AnJo   3.99
ncsB   5.36      taco   4.47      GK5.   4.01
nrfo   5.37      GK5.   4.63      taco   4.03
GK5.   5.67      nrfo   4.69      538    4.18
538    6.07      538    4.87      nrfo   4.28

lnqi returns

Code: Select all

3/06   RMSE      avg abs err             SMRE
trzu   4.22      trzu   3.39      lnqi   2.69
BaDo   4.35      lnqi   3.39      kmed   2.82
lnqi   4.40      BaDo   3.42      BaDo   2.87
RyRi   4.54      Crow   3.49      yoop   2.89
vegas  4.54     vegas   3.55      trzu   2.91
Crow   4.56      kmed   3.57      Crow   2.94
cali   4.64      yoop   3.66     vegas   2.99
sbs.   4.66      sbs.   3.70      jg34   3.03
amp5   4.67      cali   3.72      sbs.   3.13
kmed   4.72      RyRi   3.75      shad   3.17
ATCt   4.79      amp5   3.80      cali   3.20

[Crow]

Thanks for all the coverage.

[Mike G]

Code: Select all

3/10   RMSE      avg abs err             SMRE
trzu   4.02      trzu   3.30      lnqi   2.73
lnqi   4.33      lnqi   3.37      trzu   2.84
BaDo   4.33     vegas   3.47      kmed   2.85
cali   4.37      cali   3.49     vegas   2.91
vegas  4.40      BaDo   3.50      cali   2.93
RyRi   4.42      kmed   3.54      yoop   2.99
amp5   4.53      Crow   3.59      Crow   3.03
sbs.   4.54      yoop   3.60      BaDo   3.10
Crow   4.59      RyRi   3.71      sbs.   3.21
kmed   4.60      sbs.   3.72      jg34   3.23
yoop   4.62      amp5   3.74      shad   3.24
...

Code: Select all

3/13   RMSE      avg abs err             SMRE
trzu   4.26      trzu   3.48      trzu   2.95
lnqi   4.58      lnqi   3.59      lnqi   3.02
cali   4.60      cali   3.68      kmed   3.10
BaDo   4.63     vegas   3.72      Crow   3.11
vegas  4.68      BaDo   3.76      cali   3.12
RyRi   4.73      Crow   3.78     vegas   3.14
yoop   4.80      kmed   3.80      shad   3.21
amp5   4.82      yoop   3.84      yoop   3.28
Crow   4.84      RyRi   3.93      BaDo   3.31
sbs.   4.84      sbs.   3.98      amp5   3.38
kmed   4.85      amp5   3.98      jg34   3.38

[kmedved]

Thanks for the updates - I didn't have time to set up my colorful sheets this year, so it's good to see the tracking.

[Mike G]

Code: Select all

3/17   RMSE      avg abs err             SMRE
trzu   4.28      trzu   3.56      kmed   3.01
lnqi   4.59      lnqi   3.69      trzu   3.03
BaDo   4.64      cali   3.72      cali   3.16
cali   4.69      kmed   3.77      Crow   3.20
vegas  4.72      BaDo   3.77      lnqi   3.21
RyRi   4.74      yoop   3.79      yoop   3.21
yoop   4.75      Crow   3.80     vegas   3.32
kmed   4.83     vegas   3.84      BaDo   3.33
amp5   4.85      RyRi   3.99      sbs.   3.43
sbs.   4.87      amp5   4.02      shad   3.48
Crow   4.88      sbs.   4.04      ncsD   3.48
sndi   4.90      ncsD   4.04      amp5   3.52
ncsD   4.92      sndi   4.09      sndi   3.52
ATCt   5.06      shad   4.12      RyRi   3.54
shad   5.13      jg34   4.26      jg34   3.54
AnJo   5.15      ATCt   4.27      Mike   3.74
Mike   5.27      Mike   4.28      ATCt   3.77
jg34   5.35      taco   4.49      GK5.   3.81
taco   5.40      ncsB   4.52      ncsB   3.87
ncsB   5.42      AnJo   4.58      taco   3.93
nrfo   5.81      GK5.   4.68      AnJo   4.25
GK5.   5.85      538    5.11      538    4.36
538    6.23      nrfo   5.16      nrfo   4.81

Update on the best and worst looking predictions for all teams; showing all ties.

Code: Select all

.          best on          worst on
trzu   LAL,Mil,Phx,Tor       Det,Por
lnqi       GSW,Ind
cali       Atl,Sac             Mem
kmed Dal,Det,Mia,NOP,Sac   Mem,Min,Was
BaDo   Atl,Brk,Det,Hou     LAL,Mil,Phl
yoop Dal,Det,NYK,Orl,Por Brk,Chi,NOP,Sac
Crow     Bos,Mem,Tor       LAC,Phl,Phx
RyRi         Chi
amp5       Chi,OKC             Cha
sbs.     Ind,LAL,OKC
ncsD     Cha,Chi,Det       Brk,Mem,Orl
sndi     Cle,LAL,Phx         Min,Uta
shad     Chi,LAL,Tor         Brk,Sac
jg34     Chi,Ind,NYK       Mem,NOP,Orl
ATCt
Mike    Atl,Cle,SAS  Bos,Hou,LAL,Mia,Uta
taco      Chi,GSW    Brk,Cle,Ind,Orl,SAS
ncsB    Cha,Chi,Det      Brk,Mem,Orl
AnJo                         Cha
GK5. Cle,Den,Det,LAC,LAL,Uta
"                Atl,Brk,GSW,Por,SAS,Was
nrfo        SAS         Cle,Dal,Den,OKC
538      Atl,Cha,Mem,Min,Orl,Por,SAS,Was

The great 538 has no best pick, at this point.

[GK5]

6 best and 6 worst for me. I think I may have the largest difference between SMRE and RMSE as well.

[Mike G]

You are the extremest -- in (rmse - smre), you and 538 are just about tied today. Bigger number yields bigger difference.

In rmse/smre, kmed is most disparate, with a 1.61 ratio; then jg34, Crow, and GK5 in the 1.50-1.54 range. Avg is 1.41, and lowest is 1.19, by nrfo.

[permaximum]

I wanted to enter this contest but I had some serious things going on in my private life. So, hopefully I will enter the next one.

I have a feeling randomness is in play right now because most participants rely on more or less the same methods.

BTW what is SMRE? I know how it's calculated but is it scientifically proven to be reliable at picking better or accurate models? This is probably the first time I see such a variation used to pick better models.

[Mike G]

SMRE is squared mean root error, and I thought it could serve as a tie breaker, if one party takes the straight avg error while another wins in RMSE.
With tanking seeming to take the league in a big way, it may be that we benefit from shrinking the larger errors rather than squaring them. Under-performing teams are now racing to the bottom and fleeing from their expectations. Contending teams necessarily benefit from the easy wins, so keep on exceeding predictions.

And now we have 3 leaders again:

Code: Select all

3/20   RMSE      avg abs err             SMRE
trzu   4.31      lnqi   3.69      kmed   3.07
lnqi   4.53      trzu   3.71      lnqi   3.17
RyRi   4.67      kmed   3.76      ncsD   3.20
BaDo   4.69      BaDo   3.83      yoop   3.21
amp5   4.72      Crow   3.85      Crow   3.25
vegas  4.74      cali   3.86      trzu   3.31
cali   4.79      vegas  3.88      cali   3.32
sbs.   4.83      RyRi   3.93      shad   3.36
kmed   4.84      ncsD   3.93      vegas  3.36
sndi   4.87      yoop   3.93      BaDo   3.40
ncsD   4.93      amp5   3.93      amp5   3.47
yoop   4.98      shad   4.00      RyRi   3.48
ATCt   4.99      sbs.   4.04      sbs.   3.51
Crow   5.01      sndi   4.08      jg34   3.57
AnJo   5.03      ATCt   4.25      sndi   3.61
shad   5.09      jg34   4.26      Mike   3.70
Mike   5.30      Mike   4.32      ATCt   3.76
jg34   5.32      AnJo   4.44      taco   3.83
taco   5.39      taco   4.48      GK5.   3.84
ncsB   5.43      ncsB   4.51      ncsB   3.95
nrfo   5.68      GK5.   4.61      AnJo   4.06
GK5.   5.73      nrfo   4.94      538    4.45
538    6.14      538    5.08      nrfo   4.54

[Crow]

Being closest on the Celtics as of a few days ago caught my eye.

[tarrazu]

Being closest on the Celtics as of a few days ago caught my eye.

You were also closest on WAS with 46 wins. Current 538/BR projections at 49/48.4.

[Crow]

I noted the Celtics because I have a thread where I occasionally talk about them here, more than most other teams. I don't follow them closely; just poke at it. Often critical but the accurate projection suggests that I wasn't way off base on expectations for season.

[Mike G]

Did you anticipate big improvements from Thomas, Bradley, and Rozier, without much dropoff anywhere else?
Wiz also have nice stepping up from Wall, Beal, and Porter. Is it possible to predict these things?

[Crow]

My prediction that the Celtics would win a couple more games than last season (and just a couple) was more about how bad the east is, team continuity benefits, not expecting that much from Horford's arrival, and maybe not losing more games than expected by point differential as in past (by better luck or coaching improvement / team maturation) than big player gains. I did not foresee Thomas going where he has gone. That the team has only won a few more suggests most of his gains were at the expense of others and his defense. Except for PER, the other major media metrics (RPM, ws/48, BPM) think Bradley is roughly stable or declining. Rozier regressed up from ridiculously bad to just modestly weak. That part was probably predictable.

In general for predicting big improvements one could compile the years where x degree improvements occurred and see when they most commonly occur. I would think a lot of them would fall between year 3-6, with individual stats improving quicker than team impacts. And I would expect to see gains in the year when a below average to way below average coach is replaced by one somewhere above average. Expecting, but not as much. Porter has big jump in year 4, Beal yr 5. Wall improves big over a lousy year 6 but only modestly compared to year 3-5.

Celtics have 5 above average players according to BPM and WS/48. 6 by RPM. That's above average, more so for the east, but not that amazing. Re: the Celtics depth, hoopsstats.com has them 19th relative to other benches on simple plus / minus for their time, way way down from 2nd rated last season (one of the very biggest slides in league).

RPM has Thomas with 4th best team impact estimate. May be low, about right or perhaps way low. As I said in another thread, he is a case that would need and perhaps benefit from further analysis.

Almost $27 million to Horford vs. maybe a $12-15 mil. starter and the rest into a better bench? We won't know how the alternative would have done but we'll see how the chosen spending delivers in the playoffs. This season is down on RPM, BPM, PER and especially WS/48. Tends to be weaker in playoffs. Decision may look alright this season but it is not great and will more likely lol worse than improve as he ages.

The biggest bench players slide season to season according to hoopsstats? The Wizards. All the opportunity and glory to the chosen few? Seen and heard that before with Brooks. Not saying it is bad strategy. But it may not be the best in the end.

[Mike G]

Code: Select all

3/23   RMSE      avg abs err             SMRE
trzu   4.41      lnqi   3.75      lnqi   3.19
lnqi   4.62      trzu   3.83      kmed   3.24
RyRi   4.71      kmed   3.86      Crow   3.26
amp5   4.73      cali   3.90      shad   3.37
BaDo   4.77      Crow   3.91      yoop   3.39
cali   4.78      BaDo   3.93      cali   3.40
vegas  4.83      vegas  3.94      ncsD   3.44
kmed   4.88      amp5   3.94      vegas  3.44
sbs.   4.89      RyRi   3.96      amp5   3.46
ncsD   4.90      ncsD   4.00      trzu   3.47
sndi   4.93      yoop   4.02      RyRi   3.48

Code: Select all

3/26   RMSE      avg abs err             SMRE
trzu   4.56      trzu   3.94      shad   3.28
lnqi   4.79      lnqi   3.95      yoop   3.31
amp5   4.83      amp5   3.99      ncsD   3.33
RyRi   4.84      kmed   4.04      vegas  3.38
cali   4.98      vegas  4.04      kmed   3.40
AnJo   5.01      RyRi   4.05      RyRi   3.40
BaDo   5.01      cali   4.06      amp5   3.43
ncsD   5.03      ncsD   4.06      lnqi   3.45
vegas  5.06      yoop   4.08      cali   3.49
sbs.   5.07      BaDo   4.11      Crow   3.50
sndi   5.10      shad   4.12      trzu   3.50

Ampersand was middle of the pack, one week ago:

Code: Select all

3/27   RMSE      avg abs err             SMRE
trzu   4.50      amp5   3.83      shad   3.21
lnqi   4.66      lnqi   3.84      yoop   3.26
RyRi   4.68      RyRi   3.89      amp5   3.29
amp5   4.68      trzu   3.90      RyRi   3.31
ncsD   4.85      vegas  3.96      sbs.   3.32
AnJo   4.85      ncsD   3.96      vegas  3.37
cali   4.87      shad   3.99      ncsD   3.37
BaDo   4.89      kmed   3.99      lnqi   3.38
sbs.   4.92      sbs.   4.00      Crow   3.44
ATCt   4.93      BaDo   4.01      kmed   3.47
sndi   4.96      cali   4.01      trzu   3.50

Code: Select all

3/28   RMSE      avg abs err             SMRE
trzu   4.53      lnqi   3.85      yoop   3.29
lnqi   4.67      amp5   3.87      shad   3.29
RyRi   4.68      trzu   3.93      sbs.   3.34
amp5   4.70      RyRi   3.93      amp5   3.35
ncsD   4.85      kmed   3.98      lnqi   3.37
cali   4.86      BaDo   4.00      vegas  3.42
AnJo   4.88      ncsD   4.00      Crow   3.43
BaDo   4.89      vegas  4.01      ncsD   3.44
sbs.   4.94      cali   4.03      RyRi   3.44
ATCt   4.95      sbs.   4.04      kmed   3.46
kmed   4.97      shad   4.04      cali   3.55

Code: Select all

3/30   RMSE      avg abs err             SMRE
trzu   4.60      amp5   3.93      Crow   3.33
RyRi   4.80      kmed   3.96      kmed   3.35
amp5   4.80      lnqi   3.97      amp5   3.35
lnqi   4.81      trzu   3.99      vegas  3.40
cali   4.93      RyRi   3.99      RyRi   3.44
ncsD   4.94      Crow   4.02      yoop   3.44
kmed   5.00      ncsD   4.06      shad   3.47
AnJo   5.01      BaDo   4.07      lnqi   3.49
BaDo   5.01      cali   4.07      ncsD   3.52
sbs.   5.04      vegas  4.09      trzu   3.59
sndi   5.04      shad   4.15      sndi   3.60

A new front-runner every day.

Code: Select all

3/31   RMSE      avg abs err             SMRE
trzu   4.57      kmed   3.91      Crow   3.22
amp5   4.76      amp5   3.92      kmed   3.29
RyRi   4.76      lnqi   3.96      amp5   3.33
lnqi   4.78      Crow   3.97      vegas  3.43
cali   4.90      RyRi   3.99      RyRi   3.45
ncsD   4.91      trzu   3.99      lnqi   3.48
kmed   4.96      BaDo   4.02      shad   3.52
BaDo   4.99      ncsD   4.05      yoop   3.54
AnJo   4.99      cali   4.06      ncsD   3.54
sbs.   5.00      vegas  4.09      BaDo   3.57
sndi   5.00      shad   4.16      cali   3.60

Code: Select all

4/03   RMSE      avg abs err             SMRE
trzu   4.51      amp5   3.75      Crow   3.20
amp5   4.57      RyRi   3.81      shad   3.20
RyRi   4.60      trzu   3.84      amp5   3.22
lnqi   4.67      Crow   3.87      RyRi   3.26
ncsD   4.76      kmed   3.90      yoop   3.38
AnJo   4.78      lnqi   3.91      vegas  3.38
cali   4.81      BaDo   3.96      trzu   3.41
ATCt   4.83      shad   3.98      kmed   3.41
sbs.   4.86      sbs.   3.99      sbs.   3.41
kmed   4.89      vegas  4.01      BaDo   3.47
BaDo   4.89      cali   4.03      lnqi   3.47