2025-26 team win projection contest

[Crow]

Last season and this season will be least precise / highest average error for winner in last 10 years. Didn't immediately find threads from earlier years.

This season leader currently better than season, but leader error fell about a point since Jan. 15.

Did projections miss more in top third, middle, bottom or all? East / west? More error in minutes distribution or impact estimates for those who went to player level? Analysis could be done.

Bet MGM was minimizing extreme projections well at last check.

[Mike G]

quasi-monthly update on over- and under-achieving teams relative to our predictions avg.

Code: Select all

over   tm   avgA   proj      over   tm    avgA   proj
18.2  SAS   42.5   60.7     -21.5   Ind   39.3   17.8
15.8  Det   42.8   58.6     -18.5   Dal   45.6   27.1
13.2  Cha   29.9   43.1     -17.3   Sac   38.2   20.9
11.5  Phx   33.9   45.4     -13.7   Mem   42.8   29.1
11.1  Bos   43.1   54.2      -8.6   Mil   41.2   32.6
9.8   Tor   36.8   46.6      -8.3   LAC   49.3   41.0
7.6   LAL   44.0   51.6      -8.1   GSW   48.3   40.2
7.6   Mia   37.3   44.9      -4.1   Brk   25.7   21.6
6.4   Phl   37.2   43.6      -3.9   NOP   32.8   28.9
4.6   NYK   48.3   52.9      -3.8   Orl   48.9   45.1
4.1   Por   36.1   40.2      -2.9   Chi   35.7   32.8
3.0   Min   46.7   49.7      -2.2   Cle   53.0   50.8
1.3   OKC   62.7   64.0      -1.7   Was   20.5   18.8
0.8   Atl   43.4   44.2      -1.1   Den   50.6   49.5
0.4   Uta   24.0   24.4       0.0   Hou   49.7   49.7

https://www.basketball-reference.com/fr ... _prob.html

Are many of the Unders (R) apt to right their ships and get closer to expectations? If they don't, it's to the benefit of the Overs, and we can expect the overall errors to increase.
This morning, the overall avg MAE topped 8.0. It was below 7.0 as recently as Feb. 1
Last year it was a week later when we hit 8.0, en route to 8.36 by season's end.

[Crow]

6 +8 or greater overs, 7 -8 or greater unders. Balance.

[Crow]

Looks like BPM-W will be the public metrics winner this time.

[Mike G]

In the last 6 days, profound separation between the top 2 and most of the rest.

Code: Select all

diff  date:  3/21   3/27
-.05  bmgm   7.19   7.14
-.03  Ebet   7.27   7.24
.26   bpmW   7.37   7.64
.11   EExp   7.53   7.63
.27   Walr   7.70   7.97
.20   avgA   7.73   7.93
.16   Mgoo   7.74   7.90
.27   knar   7.76   8.03
.24   DRKO   7.79   8.03
.15   dtka   7.89   8.04
.13   WShr   8.07   8.20
.30   eWin   8.09   8.38
.11   perW   8.09   8.20
-.08  medi   8.44   8.36
.17   Crow   8.60   8.76
.10   25pr   8.76   8.87
.04   KPel   8.88   8.91
-.02  DQin   9.67   9.65

.13    avg   8.03   8.16

eWins "leads" in this flight from respectability. Just bad days and meh days.

[Crow]

2 betting lines doing best in worst general contest in 10 years.

What did they do that others didn't as well? More regression to mean? Better minutes distribution? Better handling of tanking and / or star resting?

Their marks team by team vs field should be reviewed. Your March 20 post with the betting lines added.

[Crow]

EBET projected 1 team 60+, 6 50+, 4 under 30.

Current BRef projection is 2 / 6 / 8

EBET underestimated the number of low win teams.

By these win levels, I was 1 / 7 / 4.


EBET had 10 from 35-45 wins. I had 11. BRef projection is for 7.

EBET had 5 from 30-34 wins. I had 2. BRef projection is for 2.

EBET had 6 from 46-50 wins. I had 5. BRef projection is for just 6.

Both EBET and I had too much in very middle and too few in very bottom. EBET also had too many in 2nd lowest tier. But I guess it was closer on the team by team detail.

[Mike G]

From Feb. 28:

Both Ebet and bmgm correctly called the dropout situation for Brk, Chi, Dal, Mem, Sac, and Uta; incorrect on Cha and Phx.
They went high on Det, NYK, Phl, Tor, Hou, Min; but also Atl, Cle, and (unresolved) Den....

Since then, Atl is 10-1, Cle 8-4, Den 9-5; Phx 6-7
AvgA has gone from .50 off the lead, to .80 out.

Those who foresaw closer East/West balance have mostly fared better than those who predicted more disparity.

Code: Select all

who     W/E   MAE
EExp   1.05   7.63
DRKO   1.06   8.03
bmgm   1.06   7.14
Ebet   1.07   7.24
Walr   1.08   7.97
eWin   1.09   8.38
Mgoo   1.10   7.90
knar   1.10   8.03
perW   1.10   8.20
Crow   1.10   8.76
KPel   1.10   8.91
WShr   1.11   8.20
medi   1.12   8.36
dtka   1.12   8.04
bpmW   1.14   7.64
DQin   1.21   9.65

Correlation here is .72
The avg prediction of W 43-39 E == 1.10 W/E
Disparity extremists >1.10 avg 8.38 MAE vs 7.73 for those <1.10

There's insignificant (.15) correlation between Avg Deviation and errors.

UPDATE Mar. 31
On Mar. 20, everyone had their worst error of the season, to date. Since then, all but one are even worse.

Code: Select all

diff  entry  3/20   3/31
.05   bmgm   7.16   7.21
.02   Ebet   7.23   7.25
.30   bpmW   7.38   7.68
.17   EExp   7.51   7.68
.36   Mgoo   7.72   8.08
.41   Walr   7.68   8.09
.38   avgA   7.70   8.09
.33   DRKO   7.80   8.12
.37   dtka   7.81   8.18
.49   knar   7.71   8.20
.23   WShr   8.05   8.29
.32   perW   8.06   8.38
.01   medi   8.43   8.45
.51   eWin   8.03   8.55
.19   Crow   8.61   8.80
.21   25pr   8.72   8.94
.21   KPel   8.83   9.04
-.11  DQin   9.65   9.54
.24    avg   7.75   7.99

Top-to-bottom difference is lowest in 6 weeks.

[Mike G]

All but the top and bottom with worst errors, per usual.

Code: Select all

.   avg err   rmse   r^2          avg err    rmse   r^2
bmgm   7.21   9.12   .53      WShr   8.41   10.93   .32
Ebet   7.22   9.29   .51      medi   8.42   10.79   .38
EExp   7.75   9.69   .46      perW   8.46   10.82   .33
bpmW   7.80   9.86   .44      eWin   8.61   10.50   .37
DRKO   8.18   9.49   .48               
Mgoo   8.20  10.18   .41      Crow   8.84   10.61   .38
avgA   8.21  10.09   .41      25pr   9.03   11.43   .27
Walr   8.23   9.87   .44      KPel   9.07   11.38   .30
knar   8.28   9.73   .46               
dtka   8.30  10.03   .42      DQin   9.56   12.45   .24

Compare to Jan. 27:

Code: Select all

.   avg err   rmse   r^2          avg err   rmse   r^2
bpmW   6.22   7.71   .52      Ebet   6.82   7.94   .54
knar   6.34   7.67   .51      dtka   6.84   8.15   .46
Walr   6.40   7.86   .50      25pr   6.88   8.99   .37
avgA   6.41   7.97   .49      Crow   6.90   8.68   .44
DRKO   6.42   7.59   .52      eWin   6.95   8.51   .42
WShr   6.48   8.48   .43      medi   7.37   8.94   .45
Mgoo   6.50   8.21   .46      perW   7.38   9.15   .35
bmgm   6.56   7.74   .56      KPel   7.63   9.21   .39
EExp   6.76   8.03   .48      DQin   8.18  10.24   .33

While current co-leaders were very middle-of-the-pack in MAE 10 weeks ago, their RMSE were top-5; and r^2 were distinctly 1 and 2. This suggests some predictive value to these other numbers, yes?

Leaders at various exponents/roots:

Code: Select all

exponents    lead
.01 - .97    Ebet
.98 - 4.3    bmgm
4.3 - 6.0    DRKO
6.1 - >>>    bmgm

UPDATE April 7 -- new leader

Code: Select all

.   avg err    rmse   r^2          avg err    rmse   r^2
Ebet   7.24    9.40   .51      knar   8.46    9.86   .46
bmgm   7.25    9.23   .53      medi   8.54   10.94   .37
EExp   7.88    9.85   .46      perW   8.58   10.91   .34
bpmW   7.90   10.01   .44      WShr   8.64   11.10   .32
DRKO   8.23    9.53   .49      eWin   8.79   10.61   .37
Walr   8.30   10.00   .44      Crow   8.88   10.71   .38
Mgoo   8.32   10.28   .41      KPel   9.17   11.52   .30
avgA   8.33   10.23   .42      25pr   9.26   11.60   .27
dtka   8.42   10.21   .42      DQin   9.64   12.60   .24

bmgm led continuously since Feb. 23 and still leads at exponent 1.03 < e < 2.90. DARKO owns the high end and Ebet the low.

[Mike G]

Ebet had 3 days on top.

Code: Select all

.   avg err    rmse   r^2          avg err    rmse   r^2
bmgm   7.22    9.25   .54      knar   8.53    9.92   .47
Ebet   7.24    9.42   .52      medi   8.60   10.99   .38
bpmW   7.94   10.06   .45      perW   8.63   10.97   .35
EExp   7.97    9.93   .46      WShr   8.74   11.18   .33
DRKO   8.23    9.57   .50      eWin   8.86   10.69   .38
Walr   8.31   10.05   .45      Crow   8.87   10.71   .39
avgA   8.37   10.28   .42      KPel   9.16   11.57   .31
Mgoo   8.39   10.34   .42      25pr   9.36   11.69   .27
dtka   8.46   10.29   .42      DQin   9.56   12.66   .24

Just 3 weeks ago, 12 others were within 1.00 of the lead; now there are 3.
Last 4 Fridays' mean absolute errors; next to the date are separation between 2nd and 3rd:

Code: Select all

3/20    .15      3/27    .40      4/03    .53      4/10    .70
bmgm   7.16      bmgm   7.14      bmgm   7.26      bmgm   7.22
Ebet   7.23      Ebet   7.24      Ebet   7.27      Ebet   7.24
bpmW   7.38                           
EExp   7.51                           
Walr   7.68      EExp   7.63      EExp   7.80         
avgA   7.70      bpmW   7.64      bpmW   7.83      bpmW   7.94
knar   7.71                                        EExp   7.97
Mgoo   7.72      Mgoo   7.90                  
DRKO   7.80      avgA   7.93      DRKO   8.23      DRKO   8.23
dtka   7.81      Walr   7.97      Mgoo   8.25         
.                DRKO   8.03      Walr   8.26         
eWin   8.03      knar   8.03      avgA   8.26         
WShr   8.05      dtka   8.04                  
perW   8.06                             

Blank lines represent gap of .20+

[Mike G]

Final.

Code: Select all

.   avg err    rmse   r^2           avg err   rmse   r^2
bmgm   7.39    9.41   .53      dtka   8.70   10.47   .41
Ebet   7.40    9.58   .51      knar   8.74   10.15   .45
EExp   8.16   10.07   .45      perW   8.76   11.09   .34
bpmW   8.23   10.24   .43      WShr   8.98   11.35   .31
DRKO   8.48    9.75   .49      eWin   9.07   10.84   .37
medi   8.58   11.13   .37      Crow   9.13   10.92   .38
Mgoo   8.60   10.50   .41      KPel   9.34   11.72   .30
avgA   8.61   10.46   .41      25pr   9.54   11.84   .26
Walr   8.64   10.27   .43      DQin   9.93   12.86   .23

In the last 12 days, medi surged from 13th to 6th and did Not end up with their worst MAE of the season.

[Crow]

MGoo up a bit, Walrus down some. BPM relatively good.

My worst ever or close to it. My very aggressive low on Heat cost almost 0.5 on average error but rest wasn't that good either.

Off by one on Thunder. BRef low estimate early on had to be undone.("Thunder projected at 54 wins. Uh huh.")
They got to do that.

I am going to look at how much different my revised projections from mid-November did, fwiw. With that set, my error on Heat shrunk to just 2 wins.

Vegas lines chosen have varied but.
should still look at their relative performance over time. Probably several wins but also some mid-pack. What is the trend?


Thanks for the tracking.

[Crow]

My revised projections on page 6 from November 16 had an average error of 5.4. I used the addition of early information pretty well.

Got within 3 wins on 11 teams. 3 direct hits, 4 just 1 off.

[Mike G]

The b-r.com projections I posted on Nov. 22, 15-17 games in, would avg 4.98 error from final wins.
Cha, Bos, SAS, LAC all ended up 10-12 wins better than that projection, Chi and Mia 10-11 worse.

Nov. 12 projections would be off by 5.86 MAE -- Chi too high by 16 -- after 10-12 games.

[Crow]

So in between those 2 dates, BRef and my revised projections were probably about equal on average error. Likely dIfferent on details.

Would be interesting to track average error for BRef model from other dates too. Even earlier, sometime in December and / or January, All-Star break... Was the trend consistently toward more predictive or not? Would more or less regressed then have done better?

I might check how Couchside and CraftedNBA projections from that time "did" compared to final results, later.