2017-18 team win projection contest

[Mike G]

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.30   6.58      GK5.   5.72   7.50
sndi   5.34   6.43            
538_   5.34   6.81      AJBk   5.82   7.12
RyRi   5.39   6.66      Mike   5.90   7.34
sbs.   5.43   6.74      ATCt   5.91   7.46
kmed   5.45   6.85      Josh   5.97   7.14
cali   5.48   6.73      Crow   6.01   7.24
                  
vegas  5.58   6.95      knar   6.11   7.48
lnqi   5.59   6.74      gold   6.11   7.25
Nath   5.62   6.88            
shad   5.63   7.11      ncs.   6.32   7.83

Entries are separated into 'tiers', defined by breaks of .10 or more.

A rags to riches story here, though far from ended. On Jan. 23, eminence hit rock bottom of the standings, 1.31 off the lead.
Here are 4-to-7 day updates of his standing since then, relative to the leader:

Code: Select all

. Date    leader  err   emin   out
Jan  23   sbs.   4.31   5.62  1.31
Jan  28   sbs.   4.14   5.03   .89
Feb   2   sbs.   4.14   5.06   .92
Feb   7   sbs.   4.01   4.96   .95
Feb  12   RyRi   4.12   4.85   .73
Feb  16   RyRi   4.20   4.71   .51
Feb  23   RyRi   4.27   4.74   .47
Mar   1   cali   4.52   5.02   .50
Mar   6   RyRi   4.85   5.17   .32
Mar  12   538.   5.21   5.39   .18
Mar  18   sndi   5.34   5.30  -.04

The last entry shows the current #2.
It's quite a precipitous worsening of the whole field. A day when you don't get worse is a good day.

[GK5]

Thanks for the additional info, Mike; you do a fantastic job moderating this contest and providing us all with interesting updates.

[Mike G]

Avg errors continue to rise.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
538_   5.46   6.92      GK5.   5.83   7.63
sndi   5.47   6.55      AJBk   5.91   7.22
RyRi   5.48   6.76      Mike   5.97   7.41
sbs.   5.50   6.83      ATCt   5.98   7.57
cali   5.51   6.82      Josh   6.07   7.25
emin   5.56   6.78      Crow   6.08   7.33
kmed   5.56   6.96      gold   6.12   7.33
lnqi   5.59   6.83      knar   6.20   7.54
vegas  5.68   7.02            
Nath   5.69   6.93            
shad   5.71   7.18      ncs.   6.49   7.91

There were times in Feb. when we were all better than the current leader errors.
Generally, there are 2 or 3 in close contention for the lead. Currently 8 are within .13 -- any of whom may lead tomorrow.

Update Mar 27, eminence now with some separation.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.45   6.78      Nath   5.78   6.99
538_   5.57   6.93      GK5.   5.96   7.76
sndi   5.64   6.65      Mike   6.09   7.55
sbs.   5.64   6.94      ATCt   6.14   7.64
cali   5.66   6.94      Josh   6.15   7.42
shad   5.69   7.27      Crow   6.15   7.35
vegas  5.70   7.12      AJBk   6.22   7.35
lnqi   5.72   6.94      gold   6.29   7.43
kmed   5.73   7.09      knar   6.30   7.56
RyRi   5.73   6.90      ncs.   6.70   8.01

The lead has changed at least 17 times since mid November:

Code: Select all

lead   date(s)  
shad  Nov 11-28
sndi  Dec 01-18
RyRi  Dec 19
sndi  Dec20
RyRi  Dec21-23
sndi  Dec24
RyRi  Dec26-Jan09
sbs   Jan10-Feb05
RyRi  Feb06-24
sbs   Feb27
cali  Mar01
RyRi  Mar03-07
538   Mar08
RyRi  Mar09
538   Mar12-17
emin  Mar18-20
538   Mar21-24
emin  Mar25-27..

I haven't recorded every day's standings, so there could be others in the lead for a day or 2.
Previous repeat-winner AJBk has slipped to 17th.

Mar. 30
Based on straight margin of victory, errors from MOV-Pythagorean wins expected:

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
cali   5.12   6.55      emin   5.53   7.00
sbs.   5.19   6.62      Nath   5.65   6.96
RyRi   5.23   6.49      GK5.   5.77   7.23
kmed   5.27   6.72      AJBk   5.78   7.03
sndi   5.29   6.45      gold   5.86   6.99
lnqi   5.29   6.54      knar   5.92   7.63
Josh   5.37   6.74      Crow   5.95   7.50
shad   5.39   7.08      ATCt   6.13   7.55
538_   5.41   6.62      Mike   6.26   7.61
vegas  5.42   6.89      ncs.   6.54   8.10

This ignores current W-L; the Cavs are a 42.7-win team, even as they've already won 45.
It's quite a different arrangement. The leaders now are middling. But the best RMSE is sndesai in both this and the b-r.com projection.
Total errors are also smaller -- Josh is .97 better in this version, and I'm .02 better.

Update Mar 31, lead change

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
538_   5.70   7.22      Nath   6.05   7.24
emin   5.71   7.00      GK5.   6.14   8.01
sbs.   5.81   7.21      ATCt   6.23   7.88
lnqi   5.81   7.17      Mike   6.32   7.86
cali   5.81   7.22      Crow   6.34   7.60
kmed   5.84   7.32      AJBk   6.38   7.61
sndi   5.84   6.95      gold   6.39   7.67
RyRi   5.92   7.17      Josh   6.44   7.76
vegas  5.93   7.39      knar   6.45   7.81
shad   5.96   7.51      ncs.   6.87   8.26

From Jan. 3 to Mar. 1, nobody was as far off as the current leaders.
Grizz keep losing, and tanking may be a wide issue. The overachievers continue to do that. Just 9 teams are within 3.0 of their projection avg.
Ranked by difference from our average guess:

Code: Select all

avg err        avg err
Mem  -14      Ind   15
GSW   -9      Phl   14
Orl   -9      Tor   11
Cha   -8      Por    9
Dal   -7      Bos    8
Phx   -7      LAL    8
SAS   -6      Hou    8
Cle   -6      NOP    4
Okl   -5      Uta    3
Atl   -4      Det    3
LAC   -4      Mia    2
Was   -3      Chi    1
Den   -3      NYK   -1
Brk   -2      Sac   -1
Mil   -2      Min   -1

Is it proper to say that relative to the Dubs, the Rockets are 17 wins better than we expected?

- Update Apr 2, after a bad Apr 1.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.75   7.01      Nath   6.09   7.30
538_   5.81   7.30      GK5.   6.12   8.09
sbs.   5.90   7.29      ATCt   6.35   8.00
cali   5.91   7.32      gold   6.41   7.74
kmed   5.91   7.36      Mike   6.42   8.02
shad   5.92   7.53      Crow   6.46   7.71
RyRi   5.94   7.23      AJBk   6.47   7.68
vegas  5.96   7.44      Josh   6.51   7.85
sndi   5.96   7.03      knar   6.61   7.93
lnqi   5.97   7.23      ncs.   6.86   8.34

Our errors are now larger than our RMSE were one month ago. Eminence leading both columns.

[eminence]

Hadn't noticed that the above post was being updated, home stretch now, hold it together boys.

[Mike G]

breaking away:

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.68   7.01      GK5.   6.10   8.08
538_   5.82   7.20      Nath   6.10   7.25
cali   5.85   7.29      ATCt   6.25   7.96
kmed   5.86   7.39      Mike   6.32   7.91
sbs.   5.91   7.27      Crow   6.40   7.66
lnqi   5.92   7.20      gold   6.44   7.73
sndi   5.92   6.97      AJBk   6.45   7.67
RyRi   5.95   7.21      knar   6.55   7.89
shad   5.97   7.58      Josh   6.60   7.83
vegas  5.97   7.46      ncs.   6.82   8.27

[Crow]

Hope a straight average gets included in the final roundup. My custom blend had too much weight on some meh to relatively bad performers this season. Apparently I gave too much credence to past performance and reputation / sophistication of method.

[Mike G]

Prize money will go to the straight winner, who may dole out parts of it to the exponential winners.
Down to the wire:

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.80   7.10      Nath   6.15   7.36
538_   5.83   7.27      GK5.   6.18   8.17
kmed   5.91   7.53      Mike   6.25   8.03
sndi   5.93   7.08      ATCt   6.32   8.10
sbs.   5.95   7.40      gold   6.44   7.84
RyRi   5.98   7.33      Crow   6.55   7.83
cali   5.98   7.45      Josh   6.64   7.89
lnqi   5.99   7.35      AJBk   6.64   7.84
shad   6.09   7.71      knar   6.76   8.04
vegas  6.10   7.61      ncs.   6.86   8.39

The top-to-bottom difference is shrinking again.

Update Apr 9 , with 1 or 2 games remaining per team.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.80   7.10      Nath   6.14   7.39
538_   5.80   7.30      GK5.   6.16   8.20
kmed   5.89   7.51      Mike   6.27   8.06
sndi   5.90   7.09      ATCt   6.30   8.12
sbs.   5.93   7.43      gold   6.46   7.90
RyRi   5.95   7.34      Crow   6.53   7.85
cali   5.96   7.48      Josh   6.62   7.88
lnqi   5.96   7.37      AJBk   6.62   7.83
vegas  6.08   7.66      knar   6.73   8.06
shad   6.11   7.74      ncs.   6.84   8.38

Update Apr 10

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
538_   5.77   7.29      shad   6.14   7.78
emin   5.81   7.09      GK5.   6.18   8.22
kmed   5.82   7.54      Mike   6.19   8.05
sndi   5.89   7.09      ATCt   6.26   8.12
sbs.   5.90   7.44      gold   6.42   7.91
RyRi   5.95   7.37      Crow   6.54   7.87
cali   5.96   7.50      Josh   6.61   7.91
lnqi   5.97   7.39      AJBk   6.67   7.89
vegas  6.04   7.66      knar   6.76   8.05
Nath   6.12   7.38      ncs.   6.81   8.38

Didn't 538 finish at the bottom last year?

[shadow]

Yeah, I think 538 was close to the bottom last year. Last year they exclusively used BPM for player projections, but this year they switched back to an RPM/BPM blend. It's supposedly 2/3 RPM and 1/3 BPM. However, when you break that down, it ends up being 2/3 box score metrics (1/3 BPM + 1/3 RPM box score component) and 1/3 RAPM (roughly half of RPM is RAPM).

[Mike G]

Every team went eminence's way last night (when possible)

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.73   7.06      shad   6.11   7.76
538_   5.74   7.21      Mike   6.14   8.01
kmed   5.80   7.52      GK5.   6.15   8.19
sndi   5.87   7.04      ATCt   6.24   8.09
sbs.   5.88   7.43      gold   6.40   7.83
RyRi   5.92   7.33      Crow   6.51   7.84
cali   5.94   7.48      Josh   6.59   7.87
lnqi   5.95   7.37      AJBk   6.64   7.89
vegas  6.02   7.66      knar   6.73   8.03
Nath   6.08   7.32      ncs.   6.78   8.35

One perhaps last look at best and worst picks:

Code: Select all

.     best                 worst
emin  Bos,Cha,Orl,Tor,Uta  
.                    Mil,Den,Por
538_  Chi,Mil,Phl,GSW    Orl,Tor
kmed  Chi,Phx,Por    Cha,Was,NOP
sndi  Cha,Den                Chi
sbs.  Mia,Sac
RyRi  Sac
cali  NYK 
lnqi  Det                    NOP 
vegas Bos,Det,Mia,NYK,LAL
Nath  Atl,Cha,Phl        Ind,Den
shad  Bos,Cha,Orl,Tor,Uta  
.                        Hou,Mem
Mike  Mia,Hou,LAL,Min,NOP,OKC
.            Brk,Det,Dal,GSW,Uta
GK5.  Cle,NYK,GSW    Phl,Phx,SAS
ATCt  Brk,NYK,Den            Cha
gold  Chi,Cle,Dal 
.            Ind,Mia,NYK,NOP,OKC
Crow  Dal,Phx           
Josh  Was            Bos,NYK,Por
AJBk  Ind                Atl,LAC
knar  Den,Mem,Phx    Det,GSW,Min
ncs.  OKC    Bos,Cle,GSW,LAC,LAL

Duplicates indicate ties.
Vegas is 5-0 on outliers but middling overall. Must be below-avg on a bunch.
Also:

Code: Select all

e =      leaders
.05   gold, Nath, Mike
.25   Mike, GK5, kmed
.65   kmed, 538, sbs
.90   538, kmed/emin
1.4   emin, sndi, 538
6.0   sndi, 538, lnqi

emin's range is .98 - 1.91, so he could take both 1 and 2; or he could finish 2nd in both ...

[eminence]

Every team went eminence's way last night (when possible)

Need one more night like that, and c'mon Jazz, grab that 3 seed.

[eminence]

Lessons I've taken from this years model:
-Trying to account for injury risk could be a fool's errand
-Weight post All-star break/trade deadline play more heavily
-Adjust extremely high win predictions down just a hair

[Mike G]

Final standings

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
emin   5.80   7.18      Mike   6.13   8.03
538_   5.83   7.24      shad   6.13   7.77
sndi   5.87   7.07      GK5.   6.29   8.24
kmed   5.95   7.61      ATCt   6.34   8.17
sbs.   5.96   7.45      gold   6.51   7.88
lnqi   5.99   7.42      Josh   6.60   7.87
cali   6.00   7.55      Crow   6.60   7.90
RyRi   6.02   7.38      AJBk   6.74   7.95
Nath   6.07   7.36      knar   6.76   8.13
vegas  6.08   7.66      ncs.   6.87   8.47

Averages are 6.23 and 7.72

[eminence]

I'd like to thank the city of Toronto for having a basketball team that made many more baskets than their opponents this year.

[Crow]

Congrats to the winner, and those close.

[Mike G]

Congrats to the winner, the other winner... and to myself for winning at the squared mean root level, or e = 0.5
By luck the final day of play delivered me 6 direct hits, or zero errors. Next was ncs with 3.

Code: Select all

SMRE / error: 0   1-2  3-5  6-9  10+

4.35   Mike   6    5    4    6    9
4.86   538_   2    4   11    6    7
4.94   kmed   2    7    7    7    7
4.98   emin   1    4   12    7    6 
4.99   sndi   2    5    9    9    5

5.01   lnqi   2    7    7    8    6
5.01   sbs.   1    7    7   10    5
5.04   vegas  1   10    3   10    6 
5.07   cali   0    9    6    8    7
5.12   RyRi   1    9    4   12    4
 
5.15   GK5.   0   11    6    5    8
5.20   ATCt   1    5    9    9    6
5.25   shad   0    9    7    8    6
5.26   Nath   1    5    9   10    5
5.52   ncs.   3    5    7    6    9

5.66   gold   1    6    5   11    7
5.87   AJBk   2    4    6    9    9
5.91   Crow   0    6    9    9    6
5.92   Josh   0    6    9    8    7
5.95   knar   0    5    9    6   10

The first 3 columns, errors up to 5, can be considered 'above avg' over the whole matrix; the last 2 columns are your bad ones. From top to bottom can be found participants with 15 on either side of the median.