2017-18 team win projection contest

[Mike G]

Taking the 2-3 games each team has played and their MOV -- not considering SOS -- we can fancifully project teams' win totals over 82 games.
These range from 2 wins for Phx up to 77/78 for the Raps and Clipps.
Using these as 25% of the projection, and our group avg. prediction for the other 75%, we get the following:

Code: Select all

GSW   60      Tor   55
Hou   58      Cle   51
SAS   55      Was   49
LAC   55      Bos   46
Okl   50      Cha   46
Por   49      Mil   43
Den   47      Mia   41
Min   46      Det   40
Uta   46      Orl   39
Mem   43      Brk   33
NOP   39      Ind   31
Dal   29      Phl   29
LAL   27      NYK   26
Sac   26      Atl   25
Phx   21      Chi   22

Relative to these projections, our average errors:

Code: Select all

kmed   3.07      Crow   3.97
cali   3.38      sndi   4.03
ATCt   3.40      538_   4.13
avg*   3.43      ncs.   4.16
RyRi   3.48      Josh   4.30
shad   3.58      emin   4.31
lnqi   3.66      knar   4.31
GK5.   3.75      AJBk   4.32
sbs.   3.92      Nath   4.33
gold   3.93      Mike   5.03

[kmedved]

Is there a spreadsheet with all the picks available? I want to run some of my own sets of rankings, and am trying to avoid duplicating the pick assembly process.

[Mike G]

You should be able to copy this into a spreadsheet.

Code: Select all

tm   Mike ncs. Josh emin shad gold 538_ RyRi AJBk sbs. sndi GK5. ATCt cali Crow kmed Nath lnqi knar
Atl   22   29   28   22   27   26   26   29   35   28   25   29   25   27   26   30   24   26   27
Bos   44   42   42   52   52   49   47   47   49   50   47   48   48   49   50   50   46   49   48
Brk   33   30   30   25   26   29   29   30   28   28   33   27   28   29   26   32   26   28   26
Cha   45   45   44   42   43   46   46   44   46   45   42   45   47   46   46   47   42   43   47
Chi   23   29   26   24   25   27   27   28   31   26   21   28   26   26   24   27   24   25   24
Cle   55   60   52   54   54   51   56   53   53   54   54   51   56   54   58   52   57   55   59
Dal   39   33   36   30   32   29   30   33   32   34   31   36   32   34   29   31   32   34   34
Den   47   47   44   52   45   47   49   46   43   46   46   50   47   45   48   46   52   46   45
Det   33   34   36   36   40   41   35   37   35   37   36   37   35   37   34   34   36   38   33
GSW   71   71   65   70   65   65   63   66   69   66   69   63   68   69   70   70   65   67   70
Hou   65   60   57   58   55   58   57   57   55   56   61   55   59   61   58   59   60   57   62
Ind   31   32   33   34   32   28   32   33   37   32   33   33   31   31   32   36   28   32   31
LAC   43   53   49   47   46   47   46   48   54   46   47   49   51   48   51   46   40   47   52
LAL   28   22   29   25   28   29   32   29   27   31   30   30   26   27   28   25   30   29   26
Mem   35   34   38   36   41   38   35   36   35   37   35   37   38   36   38   38   34   37   33
Mia   43   42   42   44   43   38   40   42   43   43   40   43   45   46   45   41   45   46   45
Mil   46   47   40   50   45   48   44   45   42   47   46   43   48   43   48   46   49   45   47
Min   47   48   51   44   45   48   50   48   47   49   50   49   50   48   45   48   46   51   51
NOP   48   45   43   40   41   38   43   41   41   41   40   43   45   39   43   39   40   39   44
NYK   29   29   34   30   31   34   30   31   33   30   31   29   28   29   32   26   32   30   30
Okl   48   48   53   49   51   56   54   50   46   52   54   47   52   52   55   49   52   51   51
Orl   33   33   33   28   33   34   37   33   34   33   33   33   33   35   33   34   34   33   31
Phl   38   34   36   35   35   38   40   36   35   36   38   30   30   35   35   33   40   37   35
Phx   31   25   28   24   29   31   31   28   26   28   29   32   25   25   23   23   26   26   23
Por   41   39   38   38   44   42   41   41   40   42   42   40   42   42   43   47   45   40   39
Sac   31   27   33   32   29   21   24   27   25   26   29   26   24   25   23   25   28   25   24
SAS   47   55   51   56   57   53   50   54   53   54   47   58   53   53   50   54   57   56   57
Tor   47   50   49   56   48   49   45   47   47   47   47   48   48   51   47   48   48   49   47
Uta   38   39   45   50   40   42   46   43   42   41   46   45   44   42   40   46   42   43   42
Was   49   48   45   47   48   48   46   48   49   48   48   45   47   46   50   50   50   48   48

Copy to Notepad, import to Excel.
Some submitted fractional wins, which would not convey.

[kmedved]

Thanks.

Here's another set of ranks, using a method suggested by Justin Kubatko some years back to regress small sample size SRS: Rest of Season SRS = (YTD_SRS*YTS_Games)/(YTS_Games+9.811).

It's still pretty noisy at this point (see Golden State for instance, who has a pretty terrible SRS), but gets meaningful pretty quickly.

[Mike G]

Here's another justforfun stab at it.
Using a hybrid (MOV+SRS)/2 in place of straight MOV in the PythW% formula, plus current wins and losses;
then using that as 50% and our avg projection as the other 50%;
we project these records:

Code: Select all

LAC   62      Tor   61
SAS   60      Cle   54
GSW   60      Orl   49
Por   58      Bos   49
Mem   51      Mia   48
Hou   50      Mil   47
Okl   46      Was   47
Uta   42      Ind   44
Den   40      Brk   40
Min   40      Cha   38
NOP   37      Det   34
LAL   25      Phl   28
Sac   21      Chi   27
Dal   20      Atl   22
Phx   16      NYK   20

In this scenario, the West averages 41.8 wins and the East 40.5
Relative to these projections, our avg errors are:

Code: Select all

kmed   6.59      GK5.   7.42
ATCt   6.88      sbs.   7.45
cali   6.99      knar   7.59
lnqi   7.11      ncs.   7.67
shad   7.14      gold   7.79
AJBk   7.21      Nath   7.94
avg*   7.31      538_   8.09
emin   7.31      sndi   8.18
RyRi   7.33      Josh   8.42
Crow   7.35      Mike   8.79

[Mike G]

Quick update; same method described above. As SRS stabilize, a large one-day improvement is seen all around.

Code: Select all

kmed   5.88      Crow   6.60
shad   5.94      AJBk   6.67
gold   6.22      emin   6.95
RyRi   6.25      ncs.   7.03
cali   6.26      538_   7.04
ATCt   6.33      knar   7.10
sbs.   6.35      Josh   7.19
lnqi   6.36      sndi   7.26
GK5.   6.39      Nath   7.31
avg*   6.43      Mike   7.97

You might think the entry known as avg* would rank higher, since half of the projections are its very values.
After 3 to 5 games, teams' rank relative to whether they're looking worse or better than our avg. prediction:

Code: Select all

Worse         Better
Min  -14      Por  16
Den  -11      Orl  16
NYK  -11      LAC  15
Dal   -8      Brk  11
Phx   -7      Tor   9
Hou   -7      Mem   8
Phl   -6      SAS   6
Uta   -6      Det   4
NOP   -5      Ind   4
Sac   -5      Mil   2
GSW   -5      Chi   1
Atl   -5      LAL   1
Cle   -3      Bos   1
Was   -3      Cha   0
Mia   -2         
Okl   -1         

[kmedved]

I slightly tweaked expected wins here by additionally regressing the impact of opposing YTD 3PT% in each's team's SRS to date. Benefits some teams like the Suns, hurts the Hornets, etc...

I also added an entrant called BetOnline (a sportsbook), with their posted Over/Unders as of October 15, and added a column showing each projected wins vs. BetOnline.

[kmedved]

For the record, I much prefer Mike G's results!

[Mike G]

So I'm dead last but would win 15 of 30 vs the betting site?
And shadow is near the top but wins only 12 ?

I see the Pinnacle line posted early-on also ranks in my 2nd division here. Is that aka Vegas?

[kmedved]

Which one is the Pinnacle line? I can add it. BetOnline may not be the best proxy for "Vegas" it occurs to me, as they often juice a lot of their lines rather than move them outright. So their Celtics line was 52.5 (-150), while other sportsbooks may have just moved the over/under down to 52.5.

[Mike G]

Pinnacle predictions were submitted by shadow, after he balanced them to 1230 total wins. It's the first submission in this thread.
He gave a prediction of 52.38 for Bos, rather identical to the 52.5 you're saying. Maybe just the difference in additive vs multiplicative adjustment?

[josecarlos]

How do you do your predictions? Which formula do you use? Which data do you use?

Thanks in advance for your answers!!!

[shadow]

Pinnacle also had some of the lines heavily juiced. So I calculated what the implied line was for each team if the juice was even on both sides. As long as you pulled the lines from BetOnline close to the start of the regular season before the market closed it's probably fine, assuming you make the adjustment for any heavily juiced sides. BetOnline usually just clones Pinnacle for the most part once Pinnacle's lines are up. You just wouldn't want to use the $250 BetOnline openers as a barometer for "Vegas" since those lines aren't very sharp and don't represent a liquid market.

Each half win was worth about 7 cents by my math. I think Pinnacle was dealing 16 cent lines, so a typical line would be Over 41 -108 / Under 41 -108. If the line was instead Over 41 -115 / Under -101, then the true line is 41.5.

[Crow]

Approaches to team projections will vary some in the details but the general approach of most is: project player performance based on past (and maybe aging, convert performance to win impacts, project minutes and sum the products. Some folks regress to mean, blend metrics, make subjective adjustments.

There may be some useful discussion to come or in past contest threads. Not everyone wants to give away the details of their approach.

The scoring methods for the contest are a kind of prediction too based on recent in season team performance. They might be easier for what you seem to want to do. Try them, see how they do, adjust them. There is no one right answer.

[josecarlos]

Approaches to team projections will vary some in the details but the general approach of most is: project player performance based on past (and maybe aging, convert performance to win impacts, project minutes and sum the products. Some folks regress to mean, blend metrics, make subjective adjustments.

There may be some useful discussion to come or in past contest threads. Not everyone wants to give away the details of their approach.

The scoring methods for the contest are a kind of prediction too based on recent in season team performance. They might be easier for what you seem to want to do. Try them, see how they do, adjust them. There is no one right answer.

Thank you Crow for your answer!!! Very interesting!!!