2017-18 team win projection contest

[Mike G]

Same method and another huge improvement by everyone:

Code: Select all

kmed   4.94      GK5.   5.68
shad   5.00     vegas   5.83
cali   5.32      emin   5.91
RyRi   5.33      538_   5.97
sbs.   5.42      sndi   6.02
lnqi   5.43      Josh   6.10
gold   5.47      ncs.   6.20
ATCt   5.48      knar   6.45
avg*   5.50      Nath   6.51
Crow   5.57      Mike   6.85
AJBk   5.58     17PyR   7.21

vegas is the aforediscussed Pinnacle line via shadow.
17PyR is last season's Pyth Wins regressed halfway to 41.
Several improve by more than 1.00 in 2 days. Several teams have decided to be more as-expected.

[Mike G]

Big shakeup as 538 surges from below-avg to the top.
Note kmedved did not get worse overnight, but everyone else looks better.
The list on the left is the avg of: our avg prediction; and the projection due to current performance (SRS+MOV)/2 record.
On the right, our group avg is ignored -- strictly current W-L and point differential applied to remaining games.

Code: Select all

(rec+pred)/2    record only
538_   4.6      ATCt    9.4
Crow   4.6      cali    9.4
cali   4.7      Crow    9.5
shad   4.8      shad    9.6
ATCt   4.9      gold    9.7
kmed   4.9      AJBk    9.8
sndi   5.0      GK5.    9.8
gold   5.0      kmed    9.9
lnqi   5.1      sndi    9.9
avg*   5.1     vegas    9.9
AJBk   5.1      538_   10.0
sbs.   5.1      sbs.   10.0
RyRi   5.2      avg*   10.2
GK5.   5.3      lnqi   10.2
emin   5.3     17PyR   10.3
Josh   5.3      Josh   10.3
knar   5.4      knar   10.3
vegas  5.5      emin   10.3
ncs.   6.0      RyRi   10.3
Nath   6.0      ncs.   10.7
Mike   6.1      Mike   10.7
17PyR  6.2      Nath   11.3

Previously, I had not entered SRS for teams whose W-L record was unchanged. But some SRS change dramatically overnight, even when the team didn't play.

UPDATE Oct 31

Code: Select all

(rec+pred)/2    record only
gold   4.9      gold   10.2
Crow   5.0      Crow   10.5
shad   5.1      kmed   10.5
kmed   5.4      shad   10.6
cali   5.4      cali   10.8
avg    5.5      AJBk   11.0
RyRi   5.5      sndi   11.0
sndi   5.6      RyRi   11.0
lnqi   5.6      lnqi   11.0
AJBk   5.6       avg   11.0
sbs.   5.6     vegas   11.1
vegas  5.6      ATCt   11.1
ATCt   5.8      sbs.   11.1
538_   5.9      538_   11.2
Josh   6.0      emin   11.3
emin   6.0      Josh   11.3
knar   6.3      knar   11.7
Nath   6.4     17PyR   11.8
GK5.   6.6      GK5.   11.9
ncs.   6.7      Nath   11.9
Mike   7.0      ncs.   12.1
17PyR  7.2      Mike   12.5

[Mike G]

Just a handful of teams are performing within the range of our predictions.
Figuring expected wins as current W + (82-W-L)*Pyth%
where Pyth% is based on (MOV+SRS)/2
... our avg errors lay out like this:

Code: Select all

avg   tm    xW    hi  low
25   Orl   58.3   37   28
16   Por   57.4   47   38
16   Ind   48.0   36   28
15   LAC   63.3   54   40
15   Mem   51.7   41   33
14   Det   49.7   41   33
12   Okl   63.0   56   46
11   Bos   59.1   52   42
9    Tor   56.9   56   45
8    Uta   51.0   50   38
8    LAL   35.8   32   22
            
3    NYK   33.6   34   26
2    Cha   46.4   47   42
2    Was   49.3   50   45
0    Phl   35.2   40   30
-1   Brk   27.5   32   25
            
-4   Phx   22.5   32   23
-4   Den   42.4   52   43
-6   NOP   35.5   48   38
-7   GSW   60.0   71   63
-9   Mil   36.4   50   40
-9   Hou   49.0   65   55
-10  Mia   33.4   46   38
-10  Atl   16.6   35   22
-11  Chi   14.7   31   21
-13  Sac   13.3   33   21
-14  Min   34.5   51   44
-14  SAS   39.5   58   47
-15  Dal   17.4   39   29
-32  Cle   23.0   60   51

So we are all 28 to 37 wins too high on the Cavs, and 21 to 30 low on the Magic, relative to their play thru 10% of the season.
Oddly, the teams with narrowest spread in our guesses -- Wiz and Hornets -- are among those performing within predictions.

[kmedved]

I'll update the BetOnline picks sometime this weekend with a better set of Vegas picks (probably the Pinnacle ones posted earlier). As mentioned, BetOnline's format makes beating the number somewhat easier since they juice the payouts a lot instead.

Remarkably little relationship between RMSE/Absolute Error rankings and record vs. BetOnline at the moment.

[nbacouchside]

Getting crushed. Smh.

[Mike G]

OKC is 4-4, and Phx is 4-5, and they're expected to win 54 and 27 respectively.
I know why this is so; it just seems funny.

[Mike G]

Now giving current records -- Wins, (MOV+SRS)/2 -- twice as much weight as 'our average prediction'.
2nd column is still averaging MOV+SRS, and we've all improved in that one.

Code: Select all

rec2+1pred     record only
gold   5.6      gold   8.8
shad   5.9      shad   9.1
Crow   6.1      538_   9.3
vegas  6.2      Crow   9.3
538_   6.3     vegas   9.4
Josh   6.3      sndi   9.5
lnqi   6.4      lnqi   9.5
cali   6.4      Josh   9.5
sndi   6.4      cali   9.6
sbs.   6.4      sbs.   9.7
AJBk   6.5      AJBk   9.8
RyRi   6.5      emin   9.8
avg-   6.6      kmed   9.8
kmed   6.6      RyRi   9.8
emin   6.8      avg-   9.8
ATCt   6.9     17PyR  10.1
Nath   7.1      ATCt  10.1
GK5.   7.1      Nath  10.2
knar   7.4      GK5.  10.3
17PyR  7.6      knar  10.6
ncs.   8.0      ncs.  11.3
Mike   8.2      Mike  11.4

I don't recall a previous contest, nor think of a reason, that our avg is doing so poorly.
It's as if a student is given 1/3 of the test answers and still gets a C-

[kmedved]

Mike correctly guessed that the "wins vs. BetOnline" column looked off. I found a fat finger error there, and now the results as much more reasonable:

[Crow]

Just to be sure Mike, you mean MOV plus SRS and not plus SOS? To average the first two?

[Mike G]

Yes. And since MOV + SOS = SRS, the formula could be stated as: (MOV*3 + SOS)/4
I just don't think either MOV or SRS can be fully trusted right now; averaging them lends some stability.

Meanwhile, I am dropping the community average factor, so we will be on a smooth continuum; with hopefully not too abrupt a shift when b-r.com Playoff Probabilities page comes online.

Code: Select all

avg error        avg error
gold   7.7      RyRi   8.8
shad   8.1      Josh   8.9
Crow   8.1      kmed   9.0
vegas  8.2      Nath   9.0
lnqi   8.6      emin   9.1
sbs.   8.6      ATCt   9.2
538_   8.7      knar   9.5
sndi   8.7      GK5.   9.7
cali   8.7      ncs.  10.1
AJBk   8.7      Mike  10.2

These errors are now a continuation of the record only column.

update Nov 7:
Errors are relative to these win projections:

Code: Select all

West  W       East  W
GSW   65      Bos   61
Okl   57      Tor   54
Hou   57      Det   52
Por   55      Was   48
LAC   55      Orl   45
Mem   51      Ind   42
Min   44      Phl   41
SAS   42      NYK   40
Uta   40      Cha   38
LAL   40      Mil   34
Den   38      Mia   34
NOP   38      Brk   27
Phx   19      Cle   27
Dal   17      Chi   24
Sac   14      Atl   20

West teams avg 42.7, East 39.3

Update Nov 8

Code: Select all

shad   7.4      sndi   8.2
gold   7.5      AJBk   8.3
Crow   7.8      kmed   8.4
vegas  7.9      Nath   8.4
sbs.   7.9      emin   8.5
cali   8.0      ATCt   8.6
lnqi   8.0      GK5.   8.9
538_   8.1      knar   9.0
Josh   8.1      Mike   9.4
RyRi   8.1      ncs.   9.5

[Mike G]

Dallas is killing me. I picked them to win 39, a full 3 wins more than the next highest and 6.3 above our avg. I wondered about this but stuck to my guns.
Last year they were the 8th oldest team at 27.6. This is one year above NBA avg. This year they are 6th oldest at 28.0
They have hardly any personnel changes from the end of last season, other than the addition of #9 draft pick Dennis Smith.

Here are some last year's and this year's stats; I used Kevin Pelton's suggested minutes, and they are shown as mpg? , as in minutes/82. Actual 2017-18 minutes/12 are shown as MPG thru 12.
Last year rates are also uncapitalized and with the question mark.

Code: Select all

Dal'18   mpg? MPG    per?   PER   ws/48?  WS/48   bpm?   BPM   rpm?
Barnes    33   35    16.3   15.2   .079   .052   -1.9   -2.8   - .3
Matthews  32   34    11.9   10.6   .060   .033     .1   -1.0     .3
Smith *   24   24    13     11.2   .07   -.074   -1     -4.3   -1.
Curry     23    0    15.5          .109           1.4            .5
Nowitzki  23   24    17.1   11.4   .084   .015   -1.3   -4.2     .3

Noel      22   17    20.4   16.3   .182   .123    3.3   - .1    1.3
Ferrell   19   28    13.2   12.3   .072   .078   - .7   - .5   -2.2
Barea     18   22    17.3   17.5   .080   .060   - .6     .3   -1.4
F-Smith   14    6     7.7    7.1   .041   .003   -1.5   -3.9   -2.2
Harris    14   12    13.8   15.4   .096   .115   - .4   - .8     .7

Mejri     11    8    14.9   21.1   .159   .181     .2    4.0    1.0
Powell     5   16    17.7    9.8   .166   .028    1.4   - .9   -1.2
Kleiber *  3    6    10     18.4   .04    .166   -3      2.2   -3.
Withey     0    2    18.8    7.3   .205  -.062    1.6  -12.9   - .9
Clavell *  0    5     8      7.5   .02    .037   -5     -2.4   -5.

totals   240  240    14.8   13.2   .092   .049   -.25   -1.6   - .3

This season's RPM are not yet available at ESPN.
Rookie ? stats are invented and don't seem too radical or influential.

A team with PER=14.8, WS/48 = .092, BPM and RPM = -1.0 to -1.3 (multiply by 4.2?) should be a nearly average team, yes?
So Curry doesn't show, and most everyone else has horrific dropoffs in play. How am I the only one to not anticipate this?

Applying last year's RPM to actual minutes, we get a team RPM of -2.6
This is still a team that should win 34 games -- not 15 to 20.

[J.E.]

What value are you using for rookies? Seems like Dennis Smith is hurting them - 46% TS% with 4.5 TO/36 - and he plays lots of minutes

[Mike G]

In the above exercise, I used .70 for DSmith; pre-season I had used .80, but it matters little.
Given the Mavs' needs, it kinda seems like they require minutes from someone who can do the things he does.

It looks like there's not much correlation between rookie performance and draft pick. I've included my guess at their rates, based entirely on pick order, shown as e484?, in which 1.00 is average. Then their rates to date (e484 as of yesterday).

Code: Select all

pic   rookie   tm    e484?  e484   PER   WS/48    BPM
1   Simmons    Phl   1.20   1.83   20.0   .147    4.2
1   Fultz      Phl   1.20    .00    3.9  -.081   -9.1
2   Ball       LAL   1.15    .78    8.9  -.022   -1.4
3   Tatum      Bos   1.10   1.35   17.6   .215    3.7
4   Jo Jackson Phx   1.05    .40    8.3  -.038   -5.5

5   Fox        Sac   1.00    .88   12.9  -.009   -3.6
6   Isaac      Orl   .95     .52   10.9   .080    -.1
7   Markkanen  Chi   .90    1.11   16.1   .131     .3
8   Ntilikina  NYK   .85     .52    8.5  -.023   -3.1
9   D Smith    Dal   .80     .79   11.2  -.074   -4.3

pic   rookie   tm    e484?  e484   PER   WS/48    BPM
10  Z Collins  Por   .75    -.87  -13.4  -.248  -12.8
11  Monk       Cha   .70     .51    9.5  -.006   -3.4
12  Kennard    Det   .65     .04    5.5  -.032   -6.0
13  Mitchell   Uta   .60     .76    9.4  -.040   -4.3
14  Adebayo    Mia   .55     .81   14.3   .141   -5.1

15  Ju Jackson Sac   .50     .13   10.4   .044   -4.0
17  Wilson     Mil   .45    -.03    6.6  -.014  -12.3
18  Leaf       Ind   .425    .28   10.2   .039   -4.1
19  J Collins  Atl   .40    1.32   20.5   .146   -1.1
22  Allen      Brk   .34     .42   10.9   .055   -1.4

23  Anunoby    Tor   .315    .78   13.5   .124    2.7
26  Swanigan   Por   .255    .64    8.4   .016   -3.1
27  Kuzma      LAL   .22    1.03   15.2   .117    -.7

I assigned .20 to all picks after 27.
Correlations to draft pick (inverted):
.09 - eWins
.01 - BPM
-.07 - PER
-.18 - WinShares
If one goes beyond the 1st round, it might get worse.

[kmedved]

Shadow has now taken the projected lead using regressed SRS as a baseline for future wins.



I feel like we're seeing bigger gaps between RMSE and absolute error than before, but that may be an early season effect.

[Mike G]

How does the Cavs' SRS of -5.0 project to 42 wins?
Orlando at +2.5 is headed to 40 wins?