Page 7 of 18
Re: 2024-25 team win projection contest
Posted: Tue Dec 10, 2024 3:46 pm
by DQuinn1575
Mike G wrote: ↑Mon Dec 09, 2024 4:37 pm
For PER and BPM, I had formulae converting to 'wins'; multiply by minutes and minimizing differences w actual league wins.
perW = (PER - 5.75)*min/4359
bpmW = (BPM + 5.25)*min/2621
WS = (WS/48)*min/49.08
Zero 'wins' are created by PER<5.75 or BPM< -5.25
These all yield 1230 wins from last year's players.
The PER win distribution was especially flat, and WS was very broad. It seems most of the better projections here are regressed toward 41-41, so I regressed WS and de-regressed eWins and PER to match the avg deviation of BPM. It turns out that these were all less than the AveDev of everyone else here; with mixed results.
Note: player eWins is by design pre-regressed when totaled for a team. A team whose players total 51 eW should win 61 games. So my eWin guesses are roughly in the middle.
Note2: I don't know how b-r.com does Win Shares. During playoff series, player WS (as WS = ws/48*min/48) never add up to 1.00 per game, and this is true in regular season as well. Overtime minutes don't explain it.
S
I used minutes from Kevin Pelton of espn.
Is anyone going to find the espn Experts' prediction? or Pelton's? I can toss out the "41 for everyone" dummy.
It's been a long time since I worked through Win Shares, but they calibrated it based on points per win. They did NOT sum up each teams to make them balance to a total, e.g. a 60 win share team could add up to 55, 60, or 65. Additionally, I don't remember for sure if they reset the points per win each season, and even if they did if the math made it so the numbers wouldnt necessarily add up to the total.
Re: 2024-25 team win projection contest
Posted: Wed Dec 11, 2024 12:55 am
by Mike G
Yes, it's based on a point differential and not on wins for a team.
But you might think that a league playing 1230 games would total 1230 WS?
Or a playoff series of 6 games (no overtimes) should total 6 and not 6.05 or 6.15?
All 4 of the prediction entries using last year's rates and Pelton's minutes came up with >1230 'wins' for this season, so they are all scaled down anyway. So it doesn't much matter that last year's players had to have their WS/48* minutes divided by 49+ rather than 48. It's just a minor quirk.
Re: 2024-25 team win projection contest
Posted: Thu Dec 12, 2024 2:34 am
by DQuinn1575
Mike G wrote: ↑Wed Dec 11, 2024 12:55 am
Yes, it's based on a point differential and not on wins for a team.
But you might think that a league playing 1230 games would total 1230 WS?
Or a playoff series of 6 games (no overtimes) should total 6 and not 6.05 or 6.15?
All 4 of the prediction entries using last year's rates and Pelton's minutes came up with >1230 'wins' for this season, so they are all scaled down anyway. So it doesn't much matter that last year's players had to have their WS/48* minutes divided by 49+ rather than 48. It's just a minor quirk.
Yeah, they did stuff like :
Calculate marginal points per win. Marginal points per win reduces to 0.32 * (league points per game) * ((team pace) / (league pace)). For the 2008-09 Cavaliers this is 0.32 * 100.0 * (88.7 / 91.7) = 30.95.
there was no calibration in total to make sure that wins shares equal to total wins. Not interested enough at the moment to see how much better one could make it if you did that.
Re: 2024-25 team win projection contest
Posted: Thu Dec 12, 2024 2:24 pm
by Mike G
We have the dreaded split leadership, with Pelton atop exponent >1.12. WS leads from .59 to .99, and TmTj below that.
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
bpmW 5.82 7.8 .51 medi 6.70 9.4 .43
WShr 5.82 8.0 .48 Crow 6.71 9.1 .46
KPel 5.86 7.5 .60 vegas 6.74 8.9 .46
TmTj 5.94 8.2 .51 eWin 6.98 9.2 .32
avgA 6.09 8.0 .49 emin 7.08 9.5 .41
DQin 6.13 8.2 .52 perW 7.59 9.8 .24
eExp 6.14 8.1 .54 24py 8.00 10.8 .33
dtka 6.56 8.1 .50 bmgm 8.54 11.1 .35
24pr 6.61 9.1 .33
We've separated into tiers. With empty lines representing ~.20 separation:
Code: Select all
. avg err
bpmW 5.82
WShr 5.82
KPel 5.86
TmTj 5.94
DQin 6.13
eExp 6.14
dtka 6.56
24pr 6.61
medi 6.70
Crow 6.71
vegas 6.74
eWin 6.98
emin 7.08
perW 7.59
24py 8.00
bmgm 8.54
Re: 2024-25 team win projection contest
Posted: Thu Dec 12, 2024 5:33 pm
by DQuinn1575
Mike G wrote: ↑Thu Dec 12, 2024 2:24 pm
We have the dreaded split leadership, with Pelton atop exponent >1.12. WS leads from .59 to .99, and TmTj below that.
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
bpmW 5.82 7.8 .51 medi 6.70 9.4 .43
WShr 5.82 8.0 .48 Crow 6.71 9.1 .46
KPel 5.86 7.5 .60 vegas 6.74 8.9 .46
TmTj 5.94 8.2 .51 eWin 6.98 9.2 .32
avgA 6.09 8.0 .49 emin 7.08 9.5 .41
DQin 6.13 8.2 .52 perW 7.59 9.8 .24
eExp 6.14 8.1 .54 24py 8.00 10.8 .33
dtka 6.56 8.1 .50 bmgm 8.54 11.1 .35
24pr 6.61 9.1 .33
We've separated into tiers. With empty lines representing ~.20 separation:
Code: Select all
. avg err
bpmW 5.82
WShr 5.82
KPel 5.86
TmTj 5.94
DQin 6.13
eExp 6.14
dtka 6.56
24pr 6.61
medi 6.70
Crow 6.71
vegas 6.74
eWin 6.98
emin 7.08
perW 7.59
24py 8.00
bmgm 8.54
HOU & MIN are 2 teams hurting me, I had them at 36 and 55 wins, and all 4 of the group ahead of me had HOU better than that and MIN worse.
Mike G wrote: ↑Thu Dec 12, 2024 2:24 pm
We have the dreaded split leadership, with Pelton atop exponent >1.12. WS leads from .59 to .99, and TmTj below that.
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
bpmW 5.82 7.8 .51 medi 6.70 9.4 .43
WShr 5.82 8.0 .48 Crow 6.71 9.1 .46
KPel 5.86 7.5 .60 vegas 6.74 8.9 .46
TmTj 5.94 8.2 .51 eWin 6.98 9.2 .32
avgA 6.09 8.0 .49 emin 7.08 9.5 .41
DQin 6.13 8.2 .52 perW 7.59 9.8 .24
eExp 6.14 8.1 .54 24py 8.00 10.8 .33
dtka 6.56 8.1 .50 bmgm 8.54 11.1 .35
24pr 6.61 9.1 .33
We've separated into tiers. With empty lines representing ~.20 separation:
Code: Select all
. avg err
bpmW 5.82
WShr 5.82
KPel 5.86
TmTj 5.94
DQin 6.13
eExp 6.14
dtka 6.56
24pr 6.61
medi 6.70
Crow 6.71
vegas 6.74
eWin 6.98
emin 7.08
perW 7.59
24py 8.00
bmgm 8.54
Really happy if I finish ahead of Vegas, and much better than 1st attempt last year.
Need HOU to do worse and MIN better, as I had HOU lower and MIN higher than any of the 4 entries in the first tier. I probably only have a shot at that tier if both of those teams change course.
Re: 2024-25 team win projection contest
Posted: Thu Dec 12, 2024 9:04 pm
by TeemoTeejay
^
I’m pretty sure you will end up in tier 1!
It’s funny, I kind of wanted to do this as a proof of concept for the metric I made, on one hand it’s doing way better than I thought it would on the other hand it’s getting smacked by winshares lol, at least at the moment
I Wonder how the BBI preseason midline predictions would look here too. Also, are the clippers at 36 or 37 wins in my projection? I saw on the list of the projections they were in 36 but with the +1 to all my stuff I thought they’d be at 37
I’m curious if the idea of having the wins closer to the midline would have change my MAE
Re: 2024-25 team win projection contest
Posted: Thu Dec 12, 2024 11:35 pm
by Mike G
Yeah you should have 37 for LAC with the adjustment. I could not seem to find why your total only increased by 29 wins.
This lowers your MAE by .04.
Everyone's MAE improves by regressing toward 41, still.
Re: 2024-25 team win projection contest
Posted: Fri Dec 13, 2024 6:22 am
by Crow
BRef is projecting Celtics to lose at about a 75% higher rate going forward than to date. Cavs 130% higher. Raptors, Hornets and Wizards to win roughly twice as frequently. Thunder to lose 50% more often. Jazz to win twice as often. Highly doubt in all these tail cases.
Re: 2024-25 team win projection contest
Posted: Fri Dec 13, 2024 7:58 am
by Crow
Who else in the media is projecting continuously updated final team win totals? Who wants to?
ESPN is:
https://www.espn.com/nba/bpi/_/view/projections
5 over 50, 5 under 30.
How different would the standings here be using ESPN instead of BRef or a blend?
What about current Vegas lines?
Or this:
https://powerrankingsguru.com/nba/win-projections.php
6 50+, 5 less than 30.
Re: 2024-25 team win projection contest
Posted: Fri Dec 13, 2024 11:32 pm
by DQuinn1575
CBS Sports also has projections that get updated, in case anyone is interested
https://www.cbssports.com/nba/standings/
Re: 2024-25 team win projection contest
Posted: Fri Dec 13, 2024 11:33 pm
by tarrazu
These do a good job.
IIRC their prior, last season, essentially mirrored the betting market:
https://www.teamrankings.com/nba/projections/standings/
*Edit -Didn't realize teamrankings was used as part of guru aggregate.
Re: 2024-25 team win projection contest
Posted: Sat Dec 14, 2024 12:42 am
by Mike G
Not too hard to plug in these various projections.
ESPN
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
KPel 4.60 6.0 .65 vegas 5.87 7.6 .54
bpmW 4.83 6.3 .60 eWin 6.06 7.8 .40
eExp 4.84 6.1 .63 Crow 6.21 8.0 .54
WShr 4.88 6.5 .57 medi 6.36 7.9 .54
avgA 5.19 6.5 .59 emin 6.68 8.6 .46
DQin 5.32 6.8 .63 perW 6.74 8.2 .34
24pr 5.44 7.6 .42 24py 7.18 9.5 .42
dtka 5.45 6.6 .61 bmgm 7.80 9.8 .46
TmTj 5.46 6.8 .61
PR Guru
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
TmTj 4.68 6.3 .68 Crow 5.44 7.2 .63
KPel 4.89 6.1 .72 medi 5.56 7.5 .60
avgA 4.90 6.3 .66 emin 5.95 7.7 .57
DQin 4.93 6.5 .68 24pr 6.03 8.0 .47
bpmW 4.99 6.5 .65 eWin 6.27 8.1 .44
dtka 5.15 6.2 .68 24py 6.77 9.2 .47
WShr 5.17 6.8 .61 perW 6.90 8.8 .36
eExp 5.27 6.8 .70 bmgm 7.31 9.4 .50
vegas 5.35 6.9 .64
CBS
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
TmTj 4.72 6.3 .68 Crow 5.38 7.2 .63
DQin 4.79 6.1 .71 vegas 5.66 7.1 .61
KPel 4.95 6.1 .73 24pr 5.67 7.8 .49
eExp 4.95 6.4 .69 emin 6.15 7.7 .57
WShr 5.01 6.7 .62 eWin 6.28 8.1 .43
bpmW 5.05 6.5 .64 24py 6.59 9.0 .49
avgA 5.10 6.3 .66 bmgm 7.01 9.2 .52
medi 5.34 7.4 .61 perW 7.23 8.8 .34
dtka 5.35 6.3 .67
Tm Rkg
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
TmTj 5.24 7.4 .65 Crow 6.18 8.0 .60
DQin 5.38 7.5 .65 eExp 6.52 8.4 .66
avgA 5.76 7.6 .63 emin 6.53 8.6 .54
bpmW 5.86 7.8 .63 eWin 7.13 9.5 .42
dtka 5.94 7.4 .65 24pr 7.17 9.5 .44
medi 6.02 8.4 .57 24py 7.54 10.1 .44
WShr 6.03 8.2 .59 bmgm 7.85 10.2 .46
vegas 6.10 7.9 .61 perW 7.85 10.1 .34
KPel 6.12 7.5 .71
If anyone reminds me, I can run these a couple more times during the season.
I would have thought that the parameters b-r.com uses -- current W-L, SRS, SOS to date, future SOS -- would be hard to beat. But for examples, the Knicks may win 44 to 51, the Wolves 42 to 48, depending who is projecting.
To-date leaders by b-r.com -- bpm and WS -- are not especially close in any of these 4.
Oh yeah, b-r.com:
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
bpmW 5.91 7.8 .51 medi 6.67 9.4 .43
WShr 5.91 8.0 .48 24pr 6.70 9.1 .33
TmTj 5.94 8.2 .51 Crow 6.72 9.1 .46
KPel 5.95 7.5 .61 vegas 6.78 8.9 .46
DQin 6.11 8.2 .53 eWin 7.02 9.2 .32
avgA 6.13 8.1 .49 emin 7.06 9.4 .42
eExp 6.21 8.1 .54 perW 7.67 9.9 .24
dtka 6.53 8.1 .51 24py 7.99 10.8 .33
. bmgm 8.50 11.1 .36
Re: 2024-25 team win projection contest
Posted: Sat Dec 14, 2024 3:05 am
by Crow
Alright, thanks for checking.
Target makes some difference. Bigger here and there.
A survey every 25 games would be nice or possibly go to a blend. Your calls.
Re: 2024-25 team win projection contest
Posted: Sat Dec 14, 2024 2:08 pm
by Mike G
Smallest errors so far for medi, dtka, avgA, and for 1 thru 15 total.
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
WShr 5.83 8.1 .49 medi 6.58 9.4 .44
bpmW 5.88 7.8 .52 24pr 6.64 9.2 .34
TmTj 5.94 8.2 .52 Crow 6.66 9.0 .47
KPel 5.95 7.5 .61 vegas 6.67 8.8 .47
avgA 6.02 8.0 .50 eWin 6.96 9.2 .33
DQin 6.06 8.3 .53 emin 7.04 9.3 .43
eExp 6.23 8.2 .56 perW 7.64 9.9 .25
dtka 6.42 8.1 .52 24py 7.90 10.8 .34
. bmgm 8.43 11.1 .36
Dec. 15:
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
WShr 5.86 8.1 .49 Crow 6.64 9.0 .47
bpmW 5.91 7.8 .52 vegas 6.64 8.8 .47
KPel 5.91 7.5 .61 24pr 6.68 9.2 .34
TmTj 5.91 8.2 .52 eWin 6.98 9.3 .33
avgA 6.01 8.0 .50 emin 7.01 9.3 .43
DQin 6.06 8.2 .53 perW 7.63 9.9 .26
eExp 6.20 8.2 .56 24py 7.89 10.8 .34
dtka 6.43 8.1 .52 bmgm 8.41 11.0 .36
medi 6.53 9.4 .44
Dec. 16 -- Season best MAE for all but PER and BPM; and a new #1
Code: Select all
. avg err rmse r^2 . avg err rmse r^2
TmTj 5.68 8.1 .53 medi 6.39 9.3 .44
WShr 5.70 8.0 .50 vegas 6.43 8.7 .48
bpmW 5.77 7.8 .53 Crow 6.56 8.9 .48
KPel 5.84 7.5 .62 24pr 6.56 9.1 .35
DQin 5.86 8.1 .54 emin 6.80 9.2 .44
avgA 5.87 8.0 .51 eWin 6.90 9.2 .34
eExp 6.09 8.2 .56 perW 7.52 9.8 .26
dtka 6.23 7.9 .53 24py 7.68 10.6 .35
. bmgm 8.21 10.9 .38
Dec. 17 -- no real change in the standings, so here are leaders at various power/root exponents:
Code: Select all
e = 4.0 e = 1.35 e = .5 e = .1
KPel 10.62 bpmW 6.50 TmTj 4.41 medi 2.50
dtka 11.03 TmTj 6.59 medi 4.62 TmTj 2.62
bpmW 11.59 KPel 6.59 WShr 4.63 dtka 3.34
eExp 11.64 WShr 6.60 DQin 4.68 DQin 3.60
WShr 11.80 DQin 6.71 bpmW 4.92 WShr 3.66
In each case, e is picked at maximum separation from the pack.
Re: 2024-25 team win projection contest
Posted: Wed Dec 18, 2024 10:35 pm
by Crow
Tight top 6. TmTj doing well as league passes 30% done. DQin close. Middle performers could still creep up.