2016-17 Team Win Projection Contest / Discussion

[Mike G]

Code: Select all

RyRi   3.47      kmed   4.11
                 sndi   4.15
amp5   3.58      ncsD   4.15
ncsB   3.65      taco   4.16
shad   3.71         
ATCt   3.71      jg34   4.29
lnqi   3.73      cali   4.29
sbs.   3.78         
BaDo   3.84      Crow   4.41
nrfo   3.87         
trzu   3.91         
Mike   3.97      yoop   4.67
vegas  3.99      GK5.   4.75

[Mike G]

Code: Select all

RyRi   3.24      nrfo   3.73         
-                BaDo   3.76      cali   4.24
amp5   3.38      Mike   3.77      Crow   4.30
lnqi   3.46      trzu   3.77         
ATCt   3.47      vegas  3.82      538    4.45
sbs.   3.50      taco   3.90      yoop   4.50
shad   3.57      kmed   3.93         
ncsB   3.59      ncsD   3.96      GK5.   4.70
-                sndi   3.98         
-                jg34   4.02         

[AcrossTheCourt]

How is 538 doing? You can choose preseason predictions:
https://projects.fivethirtyeight.com/20 ... edictions/

[Mike G]

Thanks, man. They're now inserted in the previous post.
They may do better vs their own current forecast.

They're spot on with Bos, Cha, and Ind; best guess on Bos.
They look worst (or tied) on Mem (35), Min (46), Orl (38), Phl (29), Por (46), SAS (52)

[Mike G]

Code: Select all

RyRi   3.22      ncsB   3.73      jg34   4.12
-                nrfo   3.77      ncsD   4.12
amp5   3.40      trzu   3.81         
(avg   3.41)     BaDo   3.83      cali   4.25
shad   3.45      Mike   3.85      Crow   4.33
lnqi   3.47     vegas   3.87         
sbs.   3.53      kmed   3.94      yoop   4.50
ATCt   3.60      taco   3.95      538    4.57
-                sndi   4.01      ...       
-                                 GK5.   4.81

[Mike G]

Code: Select all

RyRi   3.20      Mike   3.71         
-                BaDo   3.71      cali   4.27
-                ncsB   3.79      Crow   4.31
amp5   3.45      kmed   3.80         
shad   3.45      trzu   3.85      yoop   4.50
sbs.   3.47      taco   3.90      538    4.58
lnqi   3.47      jg34   3.95      GK5.   4.66
-                vegas  3.96         
nrfo   3.57      sndi   4.03         
ATCt   3.59      ncsD   4.05

[AcrossTheCourt]

Are you allowed to use Vegas lines in the prediction? I thought you weren't.

[Mike G]

I don't know if anyone used Vegas in their predictions. It's included here to show how well we are doing vs it.
Nobody really seems to have made clear exactly how they made their predictions. If you're in it to win it, do whatever you want. If you're testing a method, stick to that.

I've not included Vegas or 538 in our 20-predictor avg line, but that would still be in 2nd place, 0.14 worse than RyRi, if it were an actual entry.
If anyone waited to the last minute and used that as their prediction, it would still be an entry in the contest.

Meanwhile, Vegas may not have helped most of us; they're running 15th in a field of 21.

[AcrossTheCourt]

I was doing similarity measures and I noticed Ryan's entries are extremely similar to Nathan's blend, which he posted on twitter and uses Vegas numbers. They're similar to the point where it's unlikely they're his own. But I can't tell if he stole Nathan's blend with one tweak or what.

[Mike G]

With the whole field this close to being right, it's likely a lot of us have similar predictions.
Meanwhile, quite a few of us are now poised to challenge for the lead.

Code: Select all

RyRi   3.31      ncsB   3.90
-               vegas   3.90
amp5   3.44      jg34   3.92
lnqi   3.46      sndi   3.98
sbs.   3.49      taco   4.01
Mike   3.56      ncsD   4.02
nrfo   3.58         
BaDo   3.60      Crow   4.17
shad   3.64      cali   4.24
kmed   3.67         
ATCt   3.67         
-                yoop   4.44
trzu   3.80       538   4.50
-            
-                GK5.   4.66

[Mike G]

The Bucks are projecting to 45 wins, which is 10 to 17 more than our predictions. How are they doing it?

Code: Select all

eW+  Bucks    mpg   Eff%   Sco   Reb   Ast   e484       Eff%   Sco   Reb   Ast   e484
3.07  Giannis  35  .600   27.4   9.6   5.9   2.71      .554   17.0   8.0   4.0   1.43
1.71  Parker   34  .562   23.5   6.5   2.6   1.62      .528   15.7   6.0   1.8    .88
+.22   Snell   28  .544   10.8   5.0   1.6    .27  Chi .475    8.8   5.0   1.6    .16
-.28  Delly    25  .473   10.1   3.3   7.3    .44  Cle .523   10.9   3.1   6.6    .60
- -   Brogdon  23  .545   13.0   3.8   5.4    .63

+.02   Monroe  20  .558   17.9  11.6   3.8   1.74      .553   18.3  11.0   2.5   1.72
-.08  Henson   19  .534   13.5  10.0   1.9   1.01      .568   13.4   7.8   1.7   1.07
-.43 Teletovic 15  .552   15.3   5.7   1.5    .58  Phx .565   17.9   6.0   1.7   1.00
-.42  Terry    14  .491    5.9   2.4   2.7   -.20  Hou .538   10.6   2.1   2.9    .23
-.45  Beasley  13  .542   17.7   8.0   2.0   1.05  Hou .556   22.1   8.8   1.6   1.55

-.39  Plumlee   7  .518    9.2   6.1   2.7    .09      .598   11.7   9.0    .6    .92
+.14   Vaughn   5  .500   12.0   3.1    .8    .23      .397    6.9   2.9   1.1   -.18

This year on the left, last season on the right.
Sco, Reb, and Ast are per36 and standardized; mpg are min./33 (team G)

Just 2 guys may be entirely responsible for their total +3.11 eWins gain due to player improvements.
When I sub in the numbers for departed Bucks -- MCW, Bayless, Vasquez, and Ennis -- for the new guys, I get the same +3.13 total.

[Mike G]

A new front-runner, sort of.

Code: Select all

(avg   3.39)     sndi   4.06
RyRi   3.44      ncsB   4.07
lnqi   3.46      taco   4.07
BaDo   3.54      ncsD   4.07
sbs.   3.57      Crow   4.13
amp5   3.58         
Mike   3.59      cali   4.25
nrfo   3.63         
kmed   3.70      yoop   4.43
shad   3.77         
trzu   3.81         
ATCt   3.84      538    4.64
vegas  3.91         
jg34   3.97      GK5.   4.75

Regarding the use of pre-season Vegas lines in making a set of predictions: There's a pretty strong correlation between Vegas number and success to date:

Code: Select all

corr.  avg abs err      corr.  avg abs err
.984   lnqi   3.46      .959   ATCt   3.84
.983   sbs.   3.57      .957   kmed   3.70
.977    avg   3.39      .954   ncsD   4.07
.976   trzu   3.81      .951   taco   4.07
.975   RyRi   3.44      .950   ncsB   4.07
.969   sndi   4.06      .948   Crow   4.13
.968   cali   4.25      .940   shad   3.77
.966   amp5   3.58      .929   nrfo   3.63
.963   BaDo   3.54      .913    538   4.64
.962   jg34   3.97      .912   GK5.   4.75
.961   Mike   3.59      .896   yoop   4.43

I'll state unequivocally that I didn't look at Vegas until someone linked to it here in this thread.
The correlation between these Vegas similarities and avg error is -.66
Between corr. and RMSE, it's -.80 -- negative because less error is better.

Here' a closeup of just the 10 teams with which we most disagreed with Vegas. And the abs difference of our 7 most agreeable contestants:

Code: Select all

avg   tm  vegas  sbs  lqi  tzu  RR  snd  jgo  amp
5.5   Brk   20    4    3    1    6    6    7    5
4.0   Hou   43    3    2    6    3    5    1    4
3.6   Det   45    2    2    1    2    3    5    4
3.6   Ind   44    3    2    4    4    4    2    4
3.5   Mem   43    4    1    0    3    5    6    5

3.4   Sac   33    3    3    2    3    2    3    1
3.3   Min   41    1    4    3    4    3    2    4
3.1   NYK   39    4    1    2    2    1    6    2
2.9   Mil   35    2    2    0    1    1    2    1
2.6   Phl   23    1    3    0    2    5    3    5

2.4 avg /totals: 47   48   52   53   59   63   63

We were mostly or entirely higher than Vegas on Brk, Hou, Sac, NYK, Mil, and Phl; lower on Det, Ind, Mem, Min

[Mike G]

Code: Select all

lnqi    3.58        jg34    4.17
RyRi    3.66        sndi    4.18
nrfo    3.71        ncsB    4.19
BaDo    3.71        Crow    4.21
amp5    3.72        taco    4.23
Mike    3.73        ncsD    4.26
sbs.    3.74            
-                   cali    4.41
kmed    3.90            
trzu    3.96        yoop    4.64
shad    3.96            
vegas   4.00        538     4.86
ATCt    4.04        GK5.    5.00

New leader in avg error
In RMSE:

Code: Select all

RyRi    4.76        sndi    5.36
lnqi    4.85        nrfo    5.41
amp5    4.88        kmed    5.50
trzu    4.97        Crow    5.53
sbs.    5.03        cali    5.61
-                   jg34    5.63
BaDo    5.21            
vegas   5.22            
ATCt    5.22            
Mike    5.23            
ncsB    5.23        yoop    6.04
ncsD    5.23        GK5.    6.09
shad    5.29            
taco    5.29        538     6.20

[Mike G]

538 doesn't look any better vs their own projections.

Code: Select all

BaDo    3.77        jg34    4.56
-                   ncsB    4.57
-                   ncsD    4.63
lnqi    4.00            
RyRi    4.01        taco    4.77
trzu    4.03        sndi    4.83
kmed    4.04        yoop    4.83
sbs.    4.09            
Mike    4.17            
vegas   4.23            
nrfo    4.23            
amp5    4.30        538     5.23
shad    4.37            
Crow    4.40            
ATCt    4.42        GK5.    5.44
cali    4.49            

I believe these do not incorporate last night's games.
https://projects.fivethirtyeight.com/20 ... id=rrpromo
In RMSE, the host site looks even worse:

Code: Select all

trzu    5.20        cali    5.90
RyRi    5.25        taco    5.93
lnqi    5.27            
BaDo    5.33        sndi    6.04
amp5    5.42            
sbs.    5.47        jg34    6.16
ATCt    5.53            
ncsB    5.58            
vegas   5.65            
ncsD    5.65        yoop    6.46
nrfo    5.69        GK5.    6.50
Crow    5.69            
Mike    5.72            
shad    5.75        538     6.70
kmed    5.79            

[BasketDork]

Am I seeing this correctly ? This latest has me in the lead ?