Predictions 2014-2015

[permaximum]

Is 26 mpg (x82) for Kobe more optimistic than most are guessing?
His (arbitrary) 1.33 eW/484 would be lowest since his 2nd season (when he also went 26 mpg).

I bet everything I have on Kobe that he will play more than 26 mpg. I don't have to look at any statistics or metrics to see that.

As a side not, Minnesota is going to get more wins than the most of you think.

[Mike G]

26 min x 82 games = 2132 minutes
There are 39 seasons from Hall of Fame guards age 35+. Of these, 16 went more than 2130 min.
Stockton played 4 such years ; Payton and Reggie 3 each; Lenny Wilkens 2; Greer, Oscar, Drexler, and Jordan each once.
http://bkref.com/tiny/awjNP
Several of these guys didn't make it to age 36. Kobe is 36, and he had played more minutes than anyone thru age 34.

[fpliii]

Westbrook had surgery, and will miss at least one month. Assuming Durant doesn't return first, going to be pretty difficult for OKC to pile up wins.

[J.E.]

Obviously very early but here are team strength ratings for games played through Nov. 4th, and resulting expected win totals

The ratings use point differential of each game and are location/rest/OT adjusted (see viewtopic.php?f=2&t=8636) and use ridge regression (OLS leads to extremely whacky results), so there's a healthy "regression to the mean" built in. Lambda = 7.5

Code: Select all

╔══════════════════════════╦════════╦══════════╗
║           Team           ║ Rating ║ Exp Wins ║
╠══════════════════════════╬════════╬══════════╣
║ Houston Rockets          ║ 5.9    ║ 58.3     ║
║ Golden State Warriors    ║ 4.3    ║ 54       ║
║ Memphis Grizzlies        ║ 2.2    ║ 48.8     ║
║ Washington Wizards       ║ 2.2    ║ 48.6     ║
║ Brooklyn Nets            ║ 2.4    ║ 48.4     ║
║ Toronto Raptors          ║ 1.8    ║ 47.5     ║
║ Chicago Bulls            ║ 1.6    ║ 46.6     ║
║ Dallas Mavericks         ║ 1.6    ║ 46.2     ║
║ Miami Heat               ║ 1.2    ║ 46.1     ║
║ Portland Trail Blazers   ║ 1.4    ║ 45       ║
║ Milwaukee Bucks          ║ 1.1    ║ 44.6     ║
║ Sacramento Kings         ║ 0.6    ║ 43.8     ║
║ Boston Celtics           ║ 0.6    ║ 43.1     ║
║ New Orleans Pelicans     ║ 0.8    ║ 42.9     ║
║ Atlanta Hawks            ║ 0.1    ║ 41.8     ║
║ Los Angeles Clippers     ║ -0.2   ║ 40.7     ║
║ Utah Jazz                ║ 0.3    ║ 40.5     ║
║ Phoenix Suns             ║ -0.5   ║ 39.9     ║
║ Minnesota Timberwolves   ║ -0.2   ║ 39.7     ║
║ San Antonio Spurs        ║ -0.3   ║ 39.6     ║
║ Cleveland Cavaliers      ║ -1.2   ║ 37.1     ║
║ Indiana Pacers           ║ -1.2   ║ 37.1     ║
║ Denver Nuggets           ║ -1.1   ║ 37       ║
║ Charlotte Hornets        ║ -1.5   ║ 36.2     ║
║ Detroit Pistons          ║ -2.7   ║ 32.4     ║
║ New York Knicks          ║ -3.2   ║ 32.4     ║
║ Oklahoma City Thunder    ║ -3.5   ║ 29.4     ║
║ Orlando Magic            ║ -3.7   ║ 29       ║
║ Philadelphia 76ers       ║ -4.2   ║ 27.5     ║
║ Los Angeles Lakers       ║ -4.5   ║ 25.8     ║
╚══════════════════════════╩════════╩══════════╝

I'm obviously not saying the Cavs win just 37 games, that's just what their current (adjusted) point differential suggests. I think most people would agree that they have underperformed expectations though (small sample size alert)

Sacramento has been a positive surprise, going 3-1 vs GSW, POR, LAC and DEN

[AcrossTheCourt]

T
Another theory people use: Portland was worse in the second half. So what? That's not predictive of anything. Full season numbers are better. And then they beat Houston in the playoffs.

Actually, they played like a +3 team for the last 70% of the season, not just "second half". Once the Blazers stopped shooting 42% from 3pt, they fall back to their expected level. Using the Rockets series as some sort of "quality assessment" is rather odd, given the fact that overall they were actually slightly outscored and had a ton of luck in order to advance. So, yeah, based on full season data (and even something more) it is more than likely that they will not win those 50+ games again this season.

This makes no sense. They played like a +3 team the last 70%, but that's throwing out the first 30% of the season. What full season of data suggests fewer than 50 wins? You have to pick and choose to make them seem worse. If you throw out the best 1/3rds stretch of any team they look worse.

They were slightly outscored, but only slightly (by two whole points over six games), suggesting they were close in quality to Houston, a team that won 54 games with a Pyth. win total of 53. That's all I meant.

[sndesai1]

Obviously not very meaningful at this point, but just for fun, here are the RMSEs for each entry (graded against 82 * win%, 82 * pyth win% (exponent = 13.91), and actual wins + pyth win% for remaining games):



The Pyth + .500 entry is just last year's pyth win% regressed to .500, while the Vegas entry is the William Hill Sportsbook O/Us as of ~3 days before the season.

[Crow]

Appreciate the progress report.

[mystic]

T
Another theory people use: Portland was worse in the second half. So what? That's not predictive of anything. Full season numbers are better. And then they beat Houston in the playoffs.

Actually, they played like a +3 team for the last 70% of the season, not just "second half". Once the Blazers stopped shooting 42% from 3pt, they fall back to their expected level. Using the Rockets series as some sort of "quality assessment" is rather odd, given the fact that overall they were actually slightly outscored and had a ton of luck in order to advance. So, yeah, based on full season data (and even something more) it is more than likely that they will not win those 50+ games again this season.

This makes no sense. They played like a +3 team the last 70%, but that's throwing out the first 30% of the season. What full season of data suggests fewer than 50 wins? You have to pick and choose to make them seem worse. If you throw out the best 1/3rds stretch of any team they look worse.

They were slightly outscored, but only slightly (by two whole points over six games), suggesting they were close in quality to Houston, a team that won 54 games with a Pyth. win total of 53. That's all I meant.

You want to tell me, that when I base a prediction on the full season data (and something more) that this would make no sense? I guess you misread my comment in which I just pointed out that contrarily to your statement the Blazers did not just play worse in the second half of the season, but for the better part of season they played worse than in their first 30% of the games, largely explained by a difference in 3pt shooting accuracy. That's all I said. At no point did I suggest to use just part of the season for a prediction. So, sorry, if my wording might have confused you.

[Mike G]

Here's a quick and rough estimate of how much better we're doing than just assuming last year's Pythagorean wins.
These are derived from current win totals + MOV-based Pyth% for remaining games. They don't make any account for current or future SOS.

Code: Select all

crow   5.25        snd1   3.96
ncs    4.95        DrP    3.92
AJb1   4.81        atc    3.92
HDon   4.79        v-0    3.29
myst   4.68        bobb   3.09
bbs    4.53        eW     2.62
itca   4.47        AJb2   2.33
fpli   4.29        14py    .0

In this format, we can say that Crow has been twice as accurate as yours truly, at predicting changes from last year.

[Crow]

I assume you stayed loyal or pretty loyal to your metric. My approach is aimed at this challenge and again involved a metric (and prediction) blend and subjective adjustment. Different outcomes possible, but might be harder to nail it with your choice. In the end, a better metric, at least in terms of retrodiction, can be put together from optimizing the blend of metrics. It might be interesting to build the best blend for retrodicting last 5 years and see how it does this season or next five. Given time to do so.

[Mike G]

Well, b-r.com just got their 'playoff probabilities report' up. This simulates the remainder of the season, given some more reliable team strength estimates than just 5 to 7 game results. Our rankings are therefore quite different than what I put out this morning.
Avg errors:

Code: Select all

AJb1   6.2        bobb   8.0
myst   6.8        HDon   8.3
crow   7.4        ncs    8.3
v-0    7.5        snd1   8.4
bbs    7.5        itca   9.0
AJb2   7.8        DrP    9.3
atc    7.9        fpli   9.5
eW     8.0        

http://www.basketball-reference.com/fri ... f_prob.cgi

And the sqrt of the sums of the squares (RMSE) -

Code: Select all

AJb1    8.2        bobb   10.1
myst    8.6        HDon   10.3
crow    9.1        snd1   10.4
v-0     9.2        ncs    10.5
bbs     9.6        itca   10.6
AJb2    9.8        DrP    11.5
eW      9.9        fpli   11.9
atc    10.0        

[Crow]

Thanks for that comparison. JE is also doing adjusted team rankings and that might be worth a reference here as well.

[Mike G]

Well ... now that I've actually looked at b-r.com's projections, I see they're heavily regressed to the mean, that is, to .500
In fact, a prediction of 41 wins for every team would lead the field at this time, with avg 'error' of just 5.6
Squaring the errors, still a healthy lead.
Basically, "we" rank just about inversely to the standard deviation of our picks.

The Sixers are 0-7 but expected to go 28-57 the rest of the way; range is from 15 to 40 wins.

Meanwhile, a serious error got into my previous post: I had the '14 Pyth numbers scrambled, so I'm deleting that part of the post. Fact is, most of us are worse than '14py right now, according to b-r.com projections.

The b-r.com page simulates the East winning an avg of 40.1 games. Our predictions range from 37.3 (italia calcio) to 39.3 (AJb1)
Last year, they avg'd 37.5 pyth wins.

[Mike G]

Well, the b-r.com Probabilities Report still likes the Sixers to win ~18 games. http://www.basketball-reference.com/fri ... f_prob.cgi
So I'll take the perhaps unusual step of un-regressing these figures somewhat. I'll just average those projections and their simplified pythagorean number: spW = 41 + 2.6*SRS

I realize this doesn't include any info about future strength of schedule; so it's 'regressed' to zero, in a sense.
Here's the mongrel projection:

Code: Select all

W    west   spW   b-r         W   east   spW   b-r
62    GSW    69    56        62    Tor    66    57
62    Por    69    56        45    Was    44    47
62    Dal    68    55        43    Chi    42    44
56    Mem    59    53        42    Cle    43    42
54    SAS    58    50        40    Mia    39    42
53    Sac    56    49        39    Bos    39    40
51    NOP    54    47        37    Atl    35    40
48    Hou    49    47        37    Brk    36    38
47    LAC    50    45        33    Orl    30    35
45    Phx    47    44        33    Cha    31    35
38    Den    38    39        32    Ind    29    35
35    Uta    35    36        30    Mil    26    35
32    Okl    32    32        27    Det    23    30
28    LAL    27    30        25    NYK    21    29
25    Min    21    28        10    Phl     1    18

Relative to that W column, here are our rankings, relative to Pyth'14:

Code: Select all

crow   2.68        itca   1.69
bbs    2.62        eW     1.68
AJb1   2.32        bobb   1.56
myst   2.23        AJb2   1.49
ncs    2.17        fpli   1.31
atc    2.00        DrP    1.06
HDon   1.94        snd1    .87
v-0    1.74        14py    .00

And if you like squared errors, it's :

Code: Select all

crow   2.57        HDon  1.29
bbs    2.02        ncs   1.24
myst   1.94        eW    1.00
atc    1.84        fpli   .78
AJb1   1.76        DrP    .78
bobb   1.59        AJb2   .76
itca   1.41        snd1   .37
v-0    1.40        14py   .00

[bbstats]

Whoa. 2nd-to-last place?

I blame Boogie.

EDIT: Though my W totals are based on schedule so measuring them against Pyth is probably skewing it slightly?