2017-18 team win projection contest

[shadow]

Here are my current power ratings, which adjust point differential for SoS and HCA and the projected wins based on those ratings:

Code: Select all

Team  Rating Proj Wins
GSW	7.66	60.44
BOS	5.30	54.44
HOU	3.89	50.87
TOR	3.34	49.46
DET	2.89	48.32
POR	2.80	48.10
OKC	2.64	47.70
MEM	2.64	47.69
LAC	1.98	46.02
ORL	1.79	45.53
WAS	1.39	44.52
SAS	0.99	43.49
PHI	0.39	41.99
DEN	0.27	41.67
NOP	0.12	41.31
CHA	-0.58	39.53
IND	-0.59	39.49
NYK	-0.65	39.34
MIA	-0.86	38.81
MIN	-0.95	38.58
UTA	-0.97	38.53
LAL	-1.43	37.36
MIL	-1.50	37.19
CLE	-2.83	33.80
CHI	-3.15	32.98
BRK	-3.59	31.86
DAL	-4.51	29.54
ATL	-4.69	29.07
SAC	-5.11	28.01
PHX	-6.67	24.06

Here are the standings in terms of absolute error based on those projected wins:

Code: Select all

Entry Abs Error
shad	5.61
RyRi	6.01
AJBk	6.11
sbs.	6.17
gold	6.23
Josh	6.25
kmed	6.28
538_	6.29
sndi	6.48
cali	6.58
GK5.	6.60
lnqi	6.61
Crow	6.96
Nath	7.07
ncs.	7.36
ATCt	7.49
emin	7.51
Mike	7.56
knar	8.09

[kmedved]

How does the Cavs' SRS of -5.0 project to 42 wins?
Orlando at +2.5 is headed to 40 wins?

Based on research showing limited (or no) ability by teams to control their opposing 3PT%, I regress each team's opposing FT% and 3PT% to the league average (while keeping attempts as-actually-happened) and use that regressed-SRS as the baseline.

Opponents are shooting 41% from three against Cleveland, while only 31% against Orlando.

[Mike G]

That's quite a theory. Last year, teams' opponent 3FG% ranged from .324 to .382, which is slightly greater than the spread in OppTS% (.524 to .577)
Correlation to MOV was .67 for OppTS% and .65 for Opp3FG%.

Best Opp3FG% were by conference leaders GSW and Bos. In every month, they held opponents below .355 in 3FG%
Worst were non-contenders Phx and Dal, both of whom had opponents hitting > .360 in every month.

Even if future games had all teams' opponents hitting .360 from the arc, would that make up 8 ppg ?

[shadow]

FWIW, KenPom found that offenses controlled roughly 84% of the variance in 3P% in a study from 2015.

https://kenpom.com/blog/offense-vs-defe ... ercentage/

[kmedved]

I was thinking of research by @AcrossTheCourt here: https://fansided.com/2017/02/02/nylon-c ... t-defense/

The Cavs have allowed 41.4 points per 100 possessions on threes so far this year (33.5 attempts at 41.1%). If they allow 36% going forward, then that drops to 36 points, and is enough to make their adjusted net rating slightly positive.

[Mike G]

Well that's not "regressing toward average"; that would be "suddenly becoming average". After their first 2 games, the Cavs' last 10 opponents have shot better than their normal from the arc. Bulls and Pacers have hit their best 3fg% of the year, Knicks and Magic their 2nd-best.

If the Cavs 3FG defense moves halfway to average, their improvement on that end is not 5.4 Pts/100 but 2.7
And that gain could be offset by giving up something else, to put more perimeter defenders on the floor.

[Mike G]

Well yay, the b-r.com page is now running.
https://www.basketball-reference.com/fr ... _prob.html
They apply current SRS to future SOS, and then somehow regress everyone toward .500 -- Only 7 teams are outside the .450-to-.550 range by season's end.

But the focus will sharpen as time goes on. Relative to this forecast, we rank like this:

Code: Select all

avg. error         RMSE
17PyR  5.2     17PyR   6.5
shad   5.5      shad   7.1
RyRi   5.9      RyRi   7.3
AJBk   5.9      gold   7.3
sbs.   6.0      Josh   7.3
Josh   6.0      sbs.   7.4
gold   6.1      AJBk   7.5
538_   6.2      538_   7.6
sndi   6.2      avg*   7.7
vegas  6.2     vegas   7.7
avg*   6.3      lnqi   7.8
kmed   6.3      sndi   7.8
lnqi   6.4      cali   7.8
GK5.   6.5      GK5.   7.9
cali   6.5      kmed   8.0
Nath   6.8      Crow   8.4
Crow   6.8      Nath   8.5
Mike   7.1      ATCt   8.5
ATCt   7.2      Mike   8.7
ncs.   7.3      emin   9.0
emin   7.5      ncs.   9.3
knar   8.0      knar   9.5

The entry known as [2017 Pythagorean wins regressed halfway to .500] surges from last to first overnight. I'd stopped including it in my reports, since it had fallen out of 'contention' lately.
Only the Dubs are forecast to win more than 2/3 of remaining games. Only the Suns should lose more than 2/3.

[caliban]

Bulls last to 10. Just a blend of some available projections and #'s. Ranked by a Huber est this time. Set it up for MAE to take over from RMSE at the worst 10%err which now is at 12,2. Never used it before but saw that a variation was suggested at the start of the thread. An attempt to even out the worst RMSErrors in what otherwise would be a good line.

Right now shadow leading the charge against Vegas.

[Mike G]

Shadow with commanding lead, relative to b-r.com forecast.

Code: Select all

.     exp=1 exp=2          exp=1 exp=2
shad   5.2   6.7      lnqi   6.2   7.4
RyRi   5.8   7.0      GK5.   6.3   7.7
sbs.   5.8   7.1      cali   6.3   7.6
vegas  5.8   7.3      Nath   6.5   8.0
Josh   5.9   7.2      Crow   6.5   8.0
538_   5.9   7.3      Mike   6.8   8.4
gold   5.9   7.0      ATCt   6.9   8.3
AJBk   6.0   7.4      emin   7.1   8.5
sndi   6.1   7.5      ncs.   7.2   9.1
kmed   6.1   7.6      knar   7.7   9.2

[Crow]

A lot of interesting, entertaining guesses about the final results. A long way to go and the final measures will be simpler. Or can be, except in case of desire to be more complicated.

[Mike G]

I'm trying to keep it simple by using b-r.com forecast. But there's a major regression going on, and it obviously favors those who regressed heavily toward .500
My predictions ranged from 22 (Atl) to 71 (GSW); ncs had LAL w 22, and also 49 wins difference, between best and worst. We are both near the bottom right now.
The best 5 in mean absolute error had a range of 40 or less; of the bottom 6, five had range > 47. The correlation is .61

[nbacouchside]

I realized after the fact that I screwed something up with the aging curve I applied to all players, so I feel slightly less bad about my terrible placement this year.

[Crow]

It appears that the better entries still miss pretty bad (7 or more wins) on about 8 teams. Some of the lower performing entries miss that bad on 12 teams.

[nbacouchside]

It appears that the better entries still miss pretty bad (7 or more wins) on about 8 teams. Some of the lower performing entries miss that bad on 12 teams.

A lot of the "good" teams have been really disappointing and some other teams (Celtics, Magic, Pistons, Knicks) are way overperforming expectations.

[kmedved]

As you'd expect with square errors, a lot of Shadow's lead is a function of being higher on big outperformers like Detroit, Boston, and Memphis.

As an aside, there are more big overperformers (Detroit, Boston, Memphis, Orlando) relative to our average projections than big underperformers (Cleveland, Minnesota). Averaging our projections, the RMSE on teams overperforming is 8.99 vs. an RMSE of 6.47 on teams underperforming. This is contrary to what I'd have expected coming into the year, as I'd expect a few big injuries to cause a few teams to go way under, with those extra wins being picked up in a more evenly distributed way (i.e., in the abstract, I'd expect something like 18 teams to go over, and 12 teams to go under).

I will try and go look at past years to see how this actually worked out.