2019-20 team win projection contest

[Mike G]

Primary goal is always to beat vegas, secondary is to beat the average, so feeling great about how things are going so far

There are at least 2 entities known as "average".
The one I'm listing among the entries is a "contestant" whose guess for each team is the avg of what 15 apbr people guessed -- does not include vegas, 538, or 2019-regressed.
That "avg." entry briefly ranked on top of all individuals and only now slipped to 7th place.

I'm also tracking the mean/avg error of all 18 entries, which now stands at 6.30. It's the "avg avg error".
Thanks to a couple of us who are way worse than that number, 12 of you are better.
That 6.30 is also the best we've been in about a month. Peak/low was Nov. 30 (6.13), and unsurprisingly the entry known as avg. also had its best day, at 5.64

[Mike G]

liminal space has consolidated their lead, now at the top in exponents from 0.8 to 2.2
Several of us have the lead outside that range. Here are top-5's near the middle of each leader's exponent range:

Code: Select all

e = 0.20   
shad   3.71
emin   3.77
gold   3.98
Crow   4.03
.538   4.06
   
0.50   
emin   4.46
gold   4.64
Crow   4.68
bbst   4.72
lisp   4.74
   
1.50   
lisp   6.27
cali   6.35
gold   6.42
bbst   6.48
trzu   6.53
   
4.0   
trzu    9.5
cali    9.7
lisp   10.0
bbst   10.3
gold   10.4
   
10.0   
cali   14.3
trzu   14.3
Mike   14.5
Rd11   15.0
lisp   15.4
   
15.0   
Mike   15.4
cali   16.0
trzu   16.0
Rd11   16.5
lisp   17.2

Excelling in exponent <1 means you have a lot of close guesses, and the big misses don't haunt you.
At high exponent, it's all about having the least-bad miss(es).

EDIT Jan 7
A good night for everyone, as Celts, Bucks, and Thunder lost. Most would also approve of wins by Orl, Phl, Was, SAS, and Sac.

Code: Select all

.  avg err   rmse
lisp   5.33   6.9
bbst   5.41   7.2
gold   5.41   7.1
cali   5.43   7.0
Crow   5.48   7.4
avg.   5.48   7.3
emin   5.61   9.0
sndi   5.72   7.8
trzu   5.79   6.9
.538   5.82   8.0
RyRi   5.83   7.6
shad   6.01   8.1
dtka   6.35   8.1
vegas  6.38   7.8
KPel   6.44   8.2
ncs.   6.47   8.5
Rd11   7.69   9.1
Mike   8.43  10.1

Best numbers of the season for bbst, gold, cali, emin, sndi, 538, RyRi, dtka, vegas, and ncs.
trzu leads at e = 2.02+

[Mike G]

Everyone but Crow and myself showing best numbers of the season:

Code: Select all

.  avg err   rmse
lisp   5.19   6.8
bbst   5.24   7.1
gold   5.27   7.0
cali   5.37   6.9
avg.   5.40   7.3
Crow   5.41   7.3
trzu   5.43   6.7
emin   5.48   8.9
RyRi   5.62   7.5
sndi   5.63   7.7
.538   5.63   7.8
shad   5.81   8.0
vegas  6.11   7.6
dtka   6.19   8.0
ncs.   6.42   8.4
KPel   6.43   8.2
Rd11   7.62   9.2
Mike   8.34  10.1

On avg, we are .78 better than on XMas Eve: Team projections have been trending toward our expectations:

Code: Select all

LAL   62      Mil   66
LAC   55      Bos   55
Hou   54      Mia   54
Dal   53      Tor   54
Den   52      Phl   50
Uta   51      Ind   47
OKC   45      Orl   41
SAS   38      Brk   36

Por   37      Det   33
Phx   34      Chi   32
Mem   34      Was   29
Sac   34      Cha   28
NOP   33      NYK   24
Min   33      Cle   24
GSW   22      Atl   21

update Jan. 18

Code: Select all

.  avg err   rmse
lisp   5.67   7.4
Crow   5.71   7.9
bbst   5.73   7.6
emin   5.83   9.3
gold   5.85   7.5
trzu   5.90   7.3
cali   5.95   7.4
.538   5.97   8.2
avg.   6.01   7.7
RyRi   6.16   7.9
shad   6.24   8.3
sndi   6.25   8.3
vegas  6.57   8.0
dtka   6.58   8.4
ncs.   6.82   8.8
KPel   6.97   8.6
Rd11   7.86   9.5
Mike   8.72  10.4

On avg, we are .038 worse since previous update. Leader error is worst since Dec. 15

The 30 teams have sort of stratified into equal thirds: 10 that we all* guessed low on; 10 that we all* guessed high on; and 10 whose projections fall within the range of our guesses.
* indicates that there was 1 outlier guess that currently nails it.
Average over/under-achievement:

Code: Select all

low  tm    mix   tm     hi   tm
13  LAL     2   Phx     -4 * Sac
12  Dal     1   Uta     -4   Was
11  Mil     1   Den     -5 * SAS
9   Mia     1   Cle     -5   Chi
8   Tor     0   Cha     -5   Det
7   OKC    -1   Phl     -6   Por
7 * Mem    -2   Orl     -7   NOP
6 * Ind    -3   Brk     -7 * Min
6 * Bos    -4   Hou     -7   Atl
5 * LAC    -4   NYK    -26   GSW

* exceptional predictions:

Code: Select all

Mem  37   ncs
Ind  49  Crow
Bos  54  Crow
LAC  56  emin

Sac  32  dtka
SAS  37  emin
Min  31  Crow

Orlando is a special case, too. All our guesses were in a narrow range from 46.5 (Pelton) down to 41 (Crow). Their projection is now for 40.8 wins, so they belong in the 3rd group. But they're one of 7 errors within 2 of our consensus avg.

Sixers are a rather bipolar case. They're headed for 50.4, our avg guess was 51.6, and nobody guessed between 48 and 52.
11 said >52, and 4 say 48 or under.

[Mike G]

Abrupt change at the top:

Code: Select all

.  avg err   rmse   
Crow   5.67   7.9
lisp   5.84   7.5
trzu   5.88   7.3
bbst   5.95   7.7
gold   5.98   7.6
emin   6.05   9.3
cali   6.06   7.5
avg.   6.11   7.8
.538   6.18   8.4
RyRi   6.26   8.0
shad   6.37   8.4
sndi   6.43   8.4
dtka   6.54   8.4
vegas  6.71   8.1
ncs.   6.99   8.8
KPel   7.27   8.9
Rd11   8.03   9.5
Mike   8.88  10.5

In 12 games last night, lisp benefitted (relative to Crow) by a NYK loss,
Crow meanwhile was on the good side of an Ind win, Phx loss, Bos win, Min loss, Mia win.

So Crow went 5-1 vs lisp in outcomes that matter vs their predictions. And 18 of 24 outcomes had no effect.
On avg, only half of game results can matter to you, vs any other individual or the whole field -- half the time, you are favored by both teams winning or both losing, so the benefit is null.
But even at this point in the season, one day can make some big swings in the standings.

Conference imbalance has gotten more severe, with West teams now projecting to an avg 42.4 wins vs 39.6 in the East.
This is getting close to our avg of 43-39; yet our overall errors have gotten .59 worse on avg., since 2 weeks ago.

[eminence]

Always appreciate the updates Mike!

[Mike G]

Crow and I had the most outliers and hi/low predictions to start the season. Crow's are mostly working out, and mine are mostly not.
Showing + and - for highest or lowest (or tied) predictions in the field, ranked by how much he's kicking my ass.

Code: Select all

tm     Cr   proj   Mi
Mia    53 +  53  - 36
Tor    53 +  55  - 38
Bos    54 +  55  - 43
NYK    30    24  + 41
Atl    25 -  22    35
Mil    61 +  68    52
Min    30 -  30    37
Det    36    33  + 42
Cle    15 -  23  + 37
Ind    50 +  50    45
Mem    27    36  - 22
OKC    33    48  - 29
Orl    41    39    45
NOP    37 -  34    40
Chi    34    32    37
Hou    51    51    54
Phl    52    50    54
SAS    44    38    45
Brk    40    35    41

Den    57 +  51    45
Por    46    35  + 47
Dal    41    54    41
Sac    40 +  32  + 40
Was    31    29  - 28
Cha    29    26    26
Phx    32    35    36
LAL    50    62  + 55
Uta    46 -  54    51
GSW    46    20  - 41
LAC    46 -  56    52

So I have 5 guesses (bottom of list) that are looking from 2 to 6 wins closer than Crow's. And he has 5 that are 10 to 16 better than mine (at top).

Crow leads with his worst MAE since Dec. 10, I have the worst numbers of the season, and the avg of the field is worst since opening day of the current projection system (Nov 14).

Code: Select all

.    avg err   rmse
Crow    5.89    8.1
lisp    6.03    7.7
trzu    6.12    7.5
bbst    6.14    7.9
gold    6.18    7.9
emin    6.23    9.5
cali    6.27    7.7
.538    6.28    8.5
avg.    6.39    8.1
RyRi    6.52    8.2
sndi    6.61    8.6
shad    6.67    8.7
dtka    6.82    8.7
vegas   6.97    8.4
ncs.    7.18    9.0
KPel    7.46    9.1
Rd11    8.27    9.8
Mike    9.17   10.8

[Crow]

There are different kinds of analysis and analysis / response to analysis. Different people choose to do / excel at different things.

[Mike G]

I used the minutes projection that Kevin Pelton provided.
His is the other entry with the worst error of the season.
Did anyone else project those minutes onto the players?

[Crow]

Vegas vs. our average... what were their best and worst projections to date? Their motivation /context is somewhat different. Do their errors look financially motivated?

[Mike G]

Vegas has been in the bottom 1/3 all year, while "avg." has stayed in the top 1/3 and even contended for the lead.
Two weeks ago, tarrazu was cruising in the #9 spot.

Code: Select all

.  avg err   rmse
Crow   5.81   8.0
trzu   6.01   7.4
lisp   6.05   7.7
gold   6.08   7.7
bbst   6.12   7.8
emin   6.18   9.4
cali   6.22   7.6
.538   6.23   8.4
avg.   6.34   8.0
RyRi   6.35   8.1
sndi   6.47   8.5
shad   6.53   8.6
dtka   6.78   8.6
vegas  6.79   8.2
ncs.   7.15   8.9
KPel   7.45   9.0
Rd11   8.27   9.7
Mike   9.09  10.7

Besides lisp and Crow, that #2 slot has been held by cali, gold, and bbst.

Code: Select all

LAL   61      Mil   68
LAC   57      Tor   56
Uta   55      Bos   54
Dal   53      Mia   53
Den   52      Phl   51
Hou   50      Ind   49
OKC   48      Orl   38
SAS   38      Brk   35

Mem   37      Chi   32
Por   36      Det   32
NOP   35      Was   28
Phx   35      Cha   26
Sac   32      NYK   25
Min   29      Atl   23
GSW   20      Cle   22

West teams on avg project to 42.5 wins now, vs 39.5 in the East. On avg, we predicted 43-39.


Feb 5 update:

Code: Select all

.  avg err   rmse
Crow   5.80   7.7
gold   6.07   7.6
trzu   6.12   7.6
lisp   6.13   7.7
bbst   6.15   7.8
.538   6.19   8.4
RyRi   6.21   8.1
cali   6.21   7.6
emin   6.25   9.5
avg.   6.35   7.9
shad   6.45   8.5
sndi   6.46   8.4
vegas  6.72   8.2
dtka   6.81   8.6
ncs.   7.01   8.8
KPel   7.28   9.0
Rd11   8.28   9.7
Mike   9.10  10.9

This is Crow's biggest lead since Dec. 28

The #8 spot in the west has been on a rotisserie for several days.

Code: Select all

LAL   61      Mil   67
LAC   56      Tor   57
Den   53      Bos   55
Dal   52      Mia   54
Uta   52      Phl   49
Hou   51      Ind   48
OKC   49      Orl   38
Mem   38      Brk   36

Por   38      Chi   31
SAS   38      Det   30
NOP   35      Was   29
Phx   34      NYK   25
Sac   33      Cha   25
Min   28      Atl   23
GSW   21      Cle   21

The Grizz are currently 25-25 but with SRS of -2.05 and the toughest remaining schedule in the league (1.79). B-R.com expects them to go 12-20 the rest of the way.
Spurs are 22-28 with a nearly avg SRS and remaining SOS; expected to finish 16-16.
Blazers are 23-28 with -0.81 SRS and the 2nd easiest (after NOP) remaining sched; expected by b-r.com to close out 16-15 and win a playoff spot.
https://www.basketball-reference.com/fr ... _prob.html
70% chance either MIl or LAL wins the Finals. Bucks 3:1 in that matchup and better vs anyone else.

Feb6 quick update: I don't know why I find this fascinating, but here are the top 5 at various exponents; mostly where the gap is largest between #1 and #2.

Code: Select all

e=.10   
RyRi   3.59
emin   3.63
shad   3.93
Crow   4.14
bbst   4.51
   
e=.43   
emin   4.56
Crow   4.71
RyRi   4.71
shad   5.03
bbst   5.12
   
e=1   
Crow   5.75
gold   6.10
lisp   6.15
RyRi   6.15
bbst   6.15
   
e=2   
cali   7.68
gold   7.71
trzu   7.71
Crow   7.75
lisp   7.77
   
e=4.4   
trzu   10.8
cali   10.9
lisp   11.4
bbst   11.7
gold   11.7
   
e=14   
cali   16.8
trzu   16.9
Mike   16.9
Rd11   17.3
lisp   18.0
   
e=20   
Mike   17.7
cali   18.1
trzu   18.2
Rd11   18.5
lisp   19.4

Caliban leads at exponent between 1.90 and 2.95, and again from 6.6 to 15.0

[Crow]

Cavs 3-17 recently cutting my risk of being way too low on them. I may end up too low but probably not that bad.

[Mike G]

Ha. There was a day when yours was the worst Cavs' guess, but I immediately reclaimed that and am now worse by 10.
Your lead over the field is now the largest it's been since Dec.28

Code: Select all

.  avg err   rmse
Crow   5.75   7.8
gold   6.12   7.7
lisp   6.19   7.9
cali   6.20   7.7
bbst   6.21   8.0
RyRi   6.21   8.2
trzu   6.21   7.8
avg.   6.31   8.1
emin   6.31   9.7
.538   6.34   8.6
sndi   6.49   8.5
shad   6.50   8.6
vegas  6.78   8.3
dtka   6.89   8.7
ncs.   7.13   8.8
KPel   7.27   9.1
Rd11   8.29   9.8
Mike   9.12  11.0

UPDATE Feb. 23
Now the lead is smallest since Jan. 21, and a few are within one good/bad day of taking it.

Code: Select all

.  avg err   rmse
Crow   5.83   7.9
bbst   5.95   8.0
lisp   5.97   7.9
cali   5.97   7.8
gold   6.01   7.8
avg.   6.09   8.2
RyRi   6.11   8.3
.538   6.20   8.6
trzu   6.26   7.9
shad   6.39   8.7
sndi   6.42   8.6
emin   6.47   9.9
vegas  6.73   8.5
dtka   6.86   8.8
ncs.   7.04   8.8
KPel   7.05   9.2
Rd11   8.08   9.7
Mike   9.00  11.0

and this is what we are dealing with:

Code: Select all

LAL   62      Mil   68
LAC   54      Tor   58
Den   53      Bos   57
Hou   52      Mia   51
Uta   52      Phl   50
Dal   50      Ind   47
OKC   49      Brk   39
Mem   39      Orl   36
NOP   38      Was   30
SAS   37      Chi   29
Por   37      Det   28
Sac   35      Cha   27
Phx   34      NYK   25
Min   27      Atl   25
GSW   19      Cle   22

[Mike G]

Yesterday I decided to just go with the b-r.com Playoff Probabilities Report's team win projection, since that and my less informed method were converging toward identical.
Coincidentally there's a shuffle at the top. JGoldstein linked to an updating spreadsheet that incorporates other projections. Maybe that one updated yesterday, so here the MAE leaders are side by side with my previous (A) and current (B) projection errors:

Code: Select all

2-23    A      B      JG
Crow   5.83   5.74   6.04
bbst   5.95   5.82   5.98
lisp   5.97   5.85   5.95
cali   5.97   5.80   6.01
gold   6.01   5.83   5.96
RyRi   6.11   5.94   6.04
...

https://docs.google.com/spreadsheets/d/ ... 1141290980
https://www.basketball-reference.com/fr ... _prob.html
That was yesterday, and this is today, relative to b-r.com:

Code: Select all

.  avg err   rmse
cali   5.84   7.8
Crow   5.85   7.9
bbst   5.86   8.0
gold   5.88   7.8
lisp   5.89   7.9
RyRi   5.93   8.3
avg.   5.94   8.2
trzu   6.14   7.9
shad   6.22   8.7
sndi   6.34   8.6
emin   6.46   9.9
.538   6.56   9.1
vegas  6.63   8.4
dtka   6.77   8.8
ncs.   6.97   8.8
KPel   6.99   9.2
Rd11   8.03   9.8
Mike   8.83  11.0

This is as close as it has been all year, if not in any contest we've done here. Any of six contestants can be in the lead tomorrow.
Also, there is an anomaly with 538. I had entered a prediction of 45 wins for OKC. With that, their total summed to 41 wins per team. But the JGoldstein sheet shows they had 36 for OKC. Their avg wins seem now to be 40.7, and their avg error is noticeably worse.

While we normally list errors at exponents 1 and 2, I like to break it down into ranges:

Code: Select all

avg abs err   low
0.01 - 0.70   RyRi
0.71 - 0.91   bbst
0.92 - 1.13   cali
1.14 - 1.32   Crow
1.33 - 1.33   cali
1.34 - 1.72   gold
1.73 - 3.59   cali
3.60 - 5.6    trzu
5.7  - 12.0   cali
12.1 -        Mike

Occasionally one entry will lead in 2 distinct ranges, but I've not seen 3; and we have 4 for caliban.

[jgoldstein34]

Just looking at the 538 issue now, it kind of seems to me that their preseason projections using RAPTOR are changing in-season because the Thunder are not the only team where what I have down as their projection is different from what they currently list.

As for the difference between us, I suspect one of us used the RAPTOR ratings and the other using the ELO forecast? Either way, I'm going to update what I have for them to match what they currently display as RAPTOR preseason projections.

[Crow]

I expected major Thunder and Cavs trades. I assumed better lineup management by Pop.