2021-22 team win prediction contest

[wessonmo]

Here are BPI’s

edit: nvm, still working out some issues

[caliban]

Code: Select all

MIL	54,7
BRK	53,2
ATL	50,0
PHI	49,6
MIA	49,1
BOS	45,8
NYK	45,7
CHI	45,4
IND	41,9
WAS	37,5
TOR	36,7
CHO	34,0
DET	28,7
CLE	27,8
ORL	21,7
	
UTA	53,5
PHO	51,1
DAL	49,9
DEN	49,1
LAL	46,6
MEM	43,7
POR	42,7
GSW	41,2
LAC	40,7
SAS	37,0
NOP	36,9
MIN	36,4
SAC	35,5
HOU	25,4
OKC	18,7

[liminal_space]

Code: Select all

ATL	47.4
BKN	52.4
BOS	49.3
CHA	37.3
CHI	40.0
CLE	28.3
DAL	49.6
DEN	44.8
DET	27.4
GSW	41.1
HOU	24.8
IND	40.7
LAC	48.2
LAL	47.0
MEM	44.4
MIA	42.4
MIL	57.1
MIN	38.7
NOP	39.9
NYK	43.0
OKC	18.6
ORL	21.3
PHI	49.8
PHX	48.7
POR	46.3
SAC	34.8
SAS	34.3
TOR	39.0
UTA	54.9
WAS	38.8

[tarrazu]

Code: Select all

East	ATL	49.6
East	BKN	52.0
East	BOS	45.3
East	CHA	31.7
East	CHI	43.5
East	CLE	25.4
East	DET	27.7
East	IND	40.8
East	MIA	43.5
East	MIL	57.6
East	NYK	43.9
East	ORL	23.2
East	PHI	54.2
East	TOR	42.2
East	WAS	36.2
West	DAL	51.4
West	DEN	47.1
West	GSW	44.5
West	HOU	22.8
West	LAC	44.7
West	LAL	41.6
West	MEM	42.8
West	MIN	38.3
West	NOP	36.8
West	OKC	20.8
West	PHX	55.8
West	POR	46.9
West	SAC	34.1
West	SAS	32.5
West	UTA	53.3

[Crow]

UTA 57
PHX 54
DAL 48
GSW 46
LAC 45
DEN 45
LAK 46
MEM 43
SAC 39
POR 42
NOP 39
MIN 32
SAS 36
OKC 24
HOU 22

BKN 52
MIL 57
PHI 48
MIA 49
ATL 53
BOS 46
IND 43
NYK 46
CHI 43
CHA 33
TOR 39
WAS 35
CLE 24
DET 24
ORL 20

[sndesai1]

will be tracking these entries and others I've seen on twitter at this link:
https://docs.google.com/spreadsheets/d/ ... OH861z2U9/
(please correct me if i made a data entry error)

right now, the "current projection" is just the average of all but will add in updated projections as season gets going

speaking of updated projections, what are some publicly available ones? i know of 538, TeamRankings, Bk-Ref - are there others? i'd like to do some tracking of in-season projection accuracy as well after the season ends

[DarkStar48]

will be tracking these entries and others I've seen on twitter at this link:
https://docs.google.com/spreadsheets/d/ ... OH861z2U9/
(please correct me if i made a data entry error)

right now, the "current projection" is just the average of all but will add in updated projections as season gets going

speaking of updated projections, what are some publicly available ones? i know of 538, TeamRankings, Bk-Ref - are there others? i'd like to do some tracking of in-season projection accuracy as well after the season ends

I noticed that there are no Vegas totals on the sheet. Intentional?

[sndesai1]

bookmaker is actually a sportsbook

i had it readily available so i just used that

[DarkStar48]

Didn’t realize. Was unfamiliar with Bookmaker before.

Btw, for anyone who hasn’t seen these, more predictions!
-ESPN Expert Forecast (Tuesday, Oct. 19): https://www.espn.com/nba/story/_/id/323 ... l-30-teams

-BetMGM Sportsbook (Sunday, Oct. 17): https://www.vegasinsider.com/nba/odds/win-totals/

ESPN article also includes Caesars Sportsbook totals plus 538 RAPTOR predictions that were already linked in this forum

It would be interesting to compare all the publicly available oddsmakers and see which betting line is the most “reliable.”

[DarkStar48]

So, ESPN finally released the slightly-delayed win projections according to their Basketball Power Index, BPI (which is different from the Expert Forecast mentioned previously). Based on the available archived links on Internet Archive, it seems the earliest version was posted at 8:01 a.m. EST on Saturday, October 23.

URL: https://web.archive.org/web/20211023145 ... ayoff-odds

I know @Crow had left the possibility of including some late additions. Since no team had completed more than 2 games (and some were still at only 1), I wanted to suggest tracking the relative performance of the above prediction, even if not as an *official* contest participant. Mostly because BPI is one of the major publicly accessible analytic models.

[Crow]

No objections to adding whatever BPI version.

[Mike G]

Based on performance thru 3-4 games, using the SRS of b-r.com -- then regressing 50% toward 41 wins (.500 W%) -- we get these projected wins in 82 games:

Code: Select all

west   W      east   W
LAC   50      Mia   59
Mem   50      Atl   54
Uta   47      Chi   49
GSW   46      Cha   48
Dal   42      Ind   48
SAS   39      Mil   48
Den   38      Cle   46
Min   38      Was   45
LAL   37      NYK   44
Por   36      Tor   43
Sac   35      Brk   39
Hou   32      Bos   38
NOP   30      Phl   37
Phx   26      Det   35
OKC   20      Orl   30

https://www.basketball-reference.com/le ... _2022.html

Relative to these projections, our errors look like this:

Code: Select all

.   avg err   rmse
eWin   6.94   8.72
perW   7.17   8.88
cali   7.68   9.25
bpmW   7.98   9.48
lisp   8.08   9.48
trzu   8.28  10.46
shad   8.30   9.85
Crow   8.32  10.03
vegas  8.45   9.89
dtka   8.53  10.23
WShr   8.49   9.90
emin  10.26  11.77

Other than Vegas (opening post), I haven't gone looking for other/unsubmitted predictions.

[Mike G]

Wow. B-Ref has actually got their Playoff Probabilities running:
https://www.basketball-reference.com/fr ... _prob.html

Projections are heavily regressed to the mean -- an entry predicting 41 wins for every team is a full 1.00 lower MAE than anyone else.
Nevertheless, here are the standings relative to that page. Entries with low spread are heavily favored at this point in the season.

Code: Select all

.   avg err   rmse
perW   5.35   6.50
eWin   5.37   6.49
bpmW   6.73   7.66
cali   6.88   8.09
lisp   7.11   8.22
vegas  7.22   8.47
shad   7.22   8.58
Crow   7.23   8.97
trzu   7.36   9.14
dtka   7.45   8.68
WShr   7.67   8.63
emin   9.06  10.68

[Crow]

The measuring stick for the predictions is a projection of own and is challenging. Gets easier / better with time. Scorekeepers will make best efforts. Suggestions could help. What gets adopted is up to those willing to pick methods and score. Degree of agreement with / faith in accuracy of measuring stick determined by reader and may vary.

Ultimately this business will get set aside in favor of measuring against simple, final results.

So take progress reports with understanding that they are rough but probably improving estimates until the very end.

[Mike G]

Yeah so the Miami Heat have SRS of 13.7 and are 3-1. If they were really that good (all year), they would be expected to win 70-71 games.
The b-r.com projection page says they are headed to 50.4 wins; this we would expect from a team with SRS of ~4.6
So it seems they're taking just 1/3 of team SRS right now for their projections. The other 2/3 says your team is .500

(41 + 41 + 70)/3 = 50.4 (roughly)

I'm inclined to keep using them, since future strength of schedule is part of their simulation.