2021-22 team win prediction contest

[Mike G]

21re is a pretty smart but simple strategy.

You might think that pythagorean previous year wins would be a better predictor than actual wins. And I don't recall "last year" doing this well previously, but it is what it is: momentum, team culture, coaching, luck, whatev.
The "dummy" entries are mostly thrown in so that none of us will have to be in the cellar, but we've been getting trounced by one of them.

Relative to our average predictions vs current projections, over- and under-achieving teams thus far:

Code: Select all

. overachieving          underachieving
ovr  tm   avg  cur     ovr   tm   avg  cur
18   GSW   43   61     -20   NOP   41   21
18   Cle   28   46     -16   Mil   54   38
15   Was   36   52     -12   LAL   47   35
10   Chi   44   54     -11   Atl   50   39
10   Mia   45   55     -10   Hou   26   17
10   LAC   42   52      -7   Det   27   21
+7   Cha   36   43      -6   Min   36   30
+5   OKC   23   29      -4   Dal   48   43
+5   Sac   36   41      -4   Phl   50   46
+5   Tor   37   42      -4   Mem   41   38
+4   Den   46   51      -3   Brk   52   49
+1   Ind   42   43      -3   Orl   23   20
-0   Uta   53   53      -2   Bos   46   44
-1   Phx   50   50      -2   NYK   44   42
-1   SAS   35   34      -2   Por   45   43

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

Notching down the "de-regression" improves the lot of us:

Code: Select all

.   avg err   rmse         avg err  rmse
21re   6.74   8.5      lisp   7.63   9.6
Crow   6.93   9.2      bpmW   7.64   9.6
emin   6.99   9.3      trzu   7.65   9.4
avgA   7.21   9.2      2021   7.65   9.3
vegas  7.23   9.2      eWin   7.84  10.2
5.38   7.25   9.6      perW   7.91  10.4
shad   7.26   9.2      dtka   8.26  10.6
cali   7.36   9.2      4141   8.70  11.1
WShr   7.38   9.5    

update Nov. 17

Code: Select all

.   avg err   rmse         avg err   rmse
21re   6.83   8.5      2021   7.68    9.4
Crow   7.03   9.2      lisp   7.74    9.6
emin   7.11   9.3      trzu   7.85    9.6
5.38   7.35   9.7      bpmW   7.89    9.7
vegas  7.36   9.3      eWin   8.04   10.1
shad   7.43   9.2      perW   8.06   10.4
avgA   7.44   9.2      dtka   8.38   10.6
WShr   7.45   9.6      4141   8.43   10.9
cali   7.57   9.3            

update Nov 18:

Code: Select all

.   avg err   rmse          avg err    rmse
21re   6.46   8.27      cali   7.55    9.18
emin   6.97   9.09      trzu   7.61    9.37
Crow   7.14   9.11      lisp   7.64    9.51
avgA   7.22   9.09      bpmW   7.71    9.55
5.38   7.25   9.55      eWin   7.85   10.01
vegas  7.26   9.14      perW   7.92   10.33
2021   7.36   9.16      4141   8.23   10.76
shad   7.36   9.10      dtka   8.42   10.55
WShr   7.37   9.56            

update Nov 19:

Code: Select all

.   avg err   rmse          avg err   rmse
21re   6.35   8.37      lisp   7.44   9.46
emin   6.93   8.90      trzu   7.47   9.24
vegas  6.94   8.96      2021   7.48   9.20
shad   7.01   8.94      WShr   7.50   9.52
avgA   7.04   9.01      bpmW   7.73   9.52
Crow   7.05   9.03      eWin   7.82   9.96
5.38   7.18   9.47      perW   7.90  10.31
cali   7.38   9.12      4141   8.18  10.87
.                       dtka   8.32  10.45

[eminence]

Damn you 21re, keeping me from my rightful glory.

[Mike G]

Now now, let's keep it civil. You know dummies can't respond.
More major shuffling:

Code: Select all

.   avg err   rmse          avg err   rmse
21re   6.36   8.32      cali   7.20   8.86
vegas  6.64   8.83      lisp   7.30   9.24
shad   6.80   8.74      WShr   7.32   9.22
avgA   6.85   8.75      bpmW   7.44   9.31
Crow   6.86   8.82      eWin   7.61   9.72
emin   6.98   8.78      2021   7.63   9.29
5.38   7.05   9.17      perW   7.85  10.07
trzu   7.07   8.93      dtka   8.04  10.10
.                       4141   8.28  10.72

Update Nov 21:

Code: Select all

.   avg err   rmse          avg err   rmse
21re   6.67   8.50      cali   7.27   8.87
shad   6.80   8.76      WShr   7.34   9.28
vegas  6.88   8.96      bpmW   7.38   9.32
avgA   6.89   8.76      eWin   7.54   9.76
5.38   6.91   9.15      perW   7.75  10.12
trzu   6.96   8.86      2021   7.80   9.46
Crow   7.10   8.87      dtka   7.90  10.01
lisp   7.13   9.20      4141   8.36  10.86
emin   7.15   8.82            

Nov 22:

Code: Select all

.   avg err   rmse          avg err   rmse
21re   6.93   8.64      cali   7.57   9.03
Crow   7.01   8.92      WShr   7.63   9.44
shad   7.02   8.85      bpmW   7.66   9.45
trzu   7.07   8.96      eWin   7.67   9.89
emin   7.12   8.84      2021   7.77   9.52
avgA   7.12   8.88      perW   7.84  10.25
5.38   7.14   9.32      dtka   8.01  10.14
vegas  7.18   9.03      4141   8.43  11.02
lisp   7.24   9.32            

[eminence]

I'm hopeful we'll all catch those dummies in the end.

Thanks for keeping things updated!

[Mike G]

Teams are falling in line.

Code: Select all

.   avg err   rmse          avg err   rmse
Crow   6.59   8.62      cali   7.23   8.76
shad   6.64   8.50      WShr   7.43   9.32
emin   6.83   8.48      bpmW   7.50   9.35
5.38   6.86   8.95      dtka   7.65   9.82
trzu   6.86   8.64      eWin   7.65   9.92
vegas  6.87   8.73      2021   7.73   9.23
avgA   6.89   8.65      perW   7.77  10.27
lisp   6.95   8.96      4141   8.54  11.17
21re   7.02   8.57            

update 11.25 -- De-regressing the b-r.com projections only 18% now:

Code: Select all

.   avg err   rmse          avg err   rmse
Crow   6.28   8.10      cali   6.92   8.39
shad   6.36   7.94      WShr   7.04   8.83
emin   6.37   7.88      bpmW   7.09   8.74
trzu   6.40   8.11      eWin   7.14   9.29
avgA   6.52   8.07      perW   7.22   9.59
vegas  6.61   8.11      2021   7.29   8.98
21re   6.61   8.16      dtka   7.38   9.31
lisp   6.65   8.48      4141   7.99  10.75
5.38   6.72   8.55            

Top 4 slots are actual apbr individual submissions!

update 11/27

Code: Select all

.   avg err   rmse          avg err   rmse
emin   6.23   7.83      cali   7.01   8.50
Crow   6.35   8.10      bpmW   7.23   8.71
shad   6.42   8.05      eWin   7.24   9.34
trzu   6.65   8.26      WShr   7.28   8.76
avgA   6.66   8.12      perW   7.31   9.58
vegas  6.74   8.19      2021   7.42   9.07
21re   6.78   8.36      dtka   7.57   9.43
5.38   6.78   8.68      4141   8.26  10.97
lisp   6.80   8.67            

[Crow]

Between Eminence's and my projections, the Nuggets, Warriors, Clippers and Blazers might provide the final decisive margin.

Between Shadow and I, it may be decided by Spurs, Kings, Lakers and Timberwolves.


Imo, even 1 MAE of best or Vegas is pretty good. 0.1 or 0.2 from lead is very good.

[Mike G]

Dubs are projecting to ~66 wins; b-r.com gives them a range from 77 down to 50. But it will take a disaster for them to drop below 53, which was emin's outlier guess; 46 is next best.

Code: Select all

.   avg err   rmse          avg err   rmse
Crow   6.38   8.12      cali   7.11   8.48
shad   6.52   8.08      eWin   7.25   9.31
emin   6.53   7.94      perW   7.33   9.57
21re   6.69   8.28      WShr   7.33   8.78
trzu   6.73   8.28      bpmW   7.34   8.72
avgA   6.76   8.14      2021   7.41   9.11
vegas  6.78   8.22      dtka   7.53   9.46
5.38   6.80   8.74      4141   8.03  10.82
lisp   6.91   8.70            

[Crow]

Warriors great at shooting with a pretty balanced shot profile (more than just 3s). Great shot defense and defensive rebounding.

Great vet bench (of ring seekers on ring seeker pay).

3rd easiest schedule but that is a minor thing.

I do think they could get derailed by injury. Or just get beat in a tough series against right type of opponent.

But heavy favorite right now.

[Mike G]

Just 5 games last night, but wins by Mil, LAL, Mem, and Bos -- and losses for LAC, Tor, Ind -- helped most or all of us.
Only another W by GSW and L for Det worked against all.

Code: Select all

.   avg err   rmse          avg err   rmse
shad   6.24   7.99      cali   6.78   8.40
emin   6.29   7.82      eWin   6.96   9.42
Crow   6.34   8.06      WShr   6.97   8.76
trzu   6.48   8.20      bpmW   6.97   8.71
avgA   6.49   8.11      2021   7.14   9.01
vegas  6.50   8.15      perW   7.21   9.71
5.38   6.58   8.65      dtka   7.40   9.42
lisp   6.59   8.59      4141   8.23  11.03
21re   6.63   8.36            

It may be time to ditch the 41-41 entry. Any suggestions whom to put in that slot?

update Nov 30:

Code: Select all

.   avg err   rmse          avg err   rmse
Crow   5.91   7.87      WShr   6.56   8.44
shad   6.02   7.80      21re   6.62   8.23
emin   6.06   7.57      bpmW   6.70   8.36
avgA   6.17   7.82      eWin   6.71   9.05
trzu   6.23   8.03      2021   6.97   9.01
vegas  6.32   7.97      perW   7.02   9.29
cali   6.36   8.09      dtka   7.09   9.14
lisp   6.46   8.43      4141   8.06  10.82
5.38   6.49   8.44            

[Crow]

Tight race.

But yay.

[Mike G]

Code: Select all

.   avg err   rmse          avg err   rmse
shad   5.89   7.50      WShr   6.32   8.21
Crow   5.92   7.68      lisp   6.34   8.17
emin   5.98   7.38      bpmW   6.42   8.05
avgA   6.02   7.56      21re   6.42   7.95
trzu   6.06   7.82      eWin   6.58   8.79
vegas  6.18   7.67      perW   6.84   9.07
5.38   6.22   8.20      2021   6.88   8.69
cali   6.23   7.81      dtka   6.98   8.92
.                       4141   8.00  10.66

[DSMok1]

Mike G--what are you measuring against? The Basketball Reference projected win totals?

[Mike G]

Mike G--what are you measuring against? The Basketball Reference projected win totals?

Yes, almost. Earlier, they were heavily regressed toward .500; so I de-regressed them to more closely resemble their current Pythagorean performances, that is, SRS (from b-r.com) applied to future games.
Lately the b-r.com projections are not so different from the pythag, and I'm adjusting them by about 1/4 as much.

Somewhat related: I studied the last 2 years' standings and found that after 15 games, the current performances were closer to the ultimate W% than our averaged predictions. Now at ~20 games, those pre-season guesses have no bearing at all on where teams are gonna end up.

So are the Lakers really a 36-win team? They may end up better than that, or they may be worse. I think they will be closer to 36 than (our avg) 47.

[eminence]

Traditionally I've found teams and players to both stabilize pretty well by the quarter mark, so that makes sense to me. Obviously some outliers.

[Mike G]

I don't normally update every day; but when I do, it's because there are big changes from yesterday. Again:

Code: Select all

.   avg err   rmse          avg err   rmse
avgA   5.71   7.61      5.38   6.17   8.37
shad   5.72   7.64      WShr   6.17   8.29
Crow   5.83   7.82      bpmW   6.23   8.07
trzu   5.92   7.87      eWin   6.29   8.75
cali   5.93   7.85      21re   6.41   7.92
vegas  6.05   7.81      perW   6.61   8.96
emin   6.10   7.61      dtka   6.81   8.99
lisp   6.13   8.20      2021   6.82   8.85
.                       4141   7.79  10.49

Anyone on the left could take the lead overnight, apparently; or fall into the 2nd tier just as suddenly.
7th leader in 7 days.

update Dec 3, after a crazy night in the nba:

Code: Select all

.   avg err   rmse          avg err   rmse
Crow   5.66   7.82      WShr   6.38   8.42
cali   5.68   7.70      dtka   6.41   8.88
shad   5.73   7.66      21re   6.48   8.05
avgA   5.76   7.66      eWin   6.53   8.92
trzu   5.81   7.84      bpmW   6.68   8.31
lisp   5.95   8.09      2021   6.89   8.94
5.38   6.00   8.25      perW   6.98   9.13
emin   6.09   7.70      4141   7.63  10.61
vegas  6.19   7.92            

Projections for OKC and Mem changed by +/- 5 wins last night.

update Dec 5:

Code: Select all

.   avg err   rmse          avg err   rmse
cali   5.58   7.62      dtka   6.39   8.74
Crow   5.61   7.74      21re   6.46   7.91
shad   5.72   7.63      WShr   6.52   8.34
avgA   5.74   7.54      bpmW   6.67   8.18
trzu   5.85   7.87      eWin   6.71   8.71
lisp   5.97   8.00      2021   6.77   8.99
emin   6.00   7.65      perW   6.92   8.86
5.38   6.07   8.18      4141   7.66  10.35
vegas  6.16   7.85            

These are ranked by mean absolute error; by RMSE (squaring the errors, summing them, and taking the 1/2 root of that sum) avgA leads.
A number of entries lead at different exponents:

Code: Select all

exp:    From   To
Crow    .10 -  .85
cali    .86 - 1.64
avgA   1.65 - 2.43
emin   2.44 - 3.33
21re   3.34 - 6.4
bpmW   6.5  - -etc

Crow dominates the low end by virtue of having many close guesses; a bad one (Cle) doesn't matter so much here.
High exponent leader is entirely the result of having the smallest worst guess: bpm's worst error is 20.1 for the Dubs. Everyone else has at least one bigger than that; and if the Cavs fade, this will change.