Forecast Model

[Johndoe804]

Hey,

I'm new to this board, and this is my first topic. I'm pretty interested in the developments being made in analyzing basketball players and teams using advanced statistics. Apparently, this board is the place to be for discussing the topic.

At any rate, I'd like to develop a model for my econometrics class that takes relevant statistics pertaining to the performance of college basketball players and places some monetary value on those players based on those statistics, essentially forecasting what sort of salary a player could expect in the NBA. Of course, that is made difficult by upcoming changes to the league's CBA. But I'm also concerned with which college statistics translate best to the NBA, as well as with how exactly to assign a monetary value to players coming into the NBA. I've got some ideas, but I'd like to hear the opinions and ideas of people with more experience in this field.

Any help would be very much appreciated.

[Crow]

Hey.

There is one recovered thread (with links from it) of note about NCAA-NBA translation:
http://sonicscentral.com/apbrmetrics/vi ... ?f=2&t=181

There is also apparently a recent article on the topic in the pay section of basketbalprospectus.com.


There was this old thread on calculating value http://sonicscentral.com/apbrmetrics/vi ... p?f=2&t=87

You can also look at stuff by DSMok1 about how he converted stats to value. Here and at his site http://godismyjudgeok.com/DStats
I don't immediately see the exact right links.

bbstats may have some useful stuff at his site or ideas about college to NBA and perhaps valuation.

Kevin Pelton had some old articles at Basketball Prospectus that converted stats to value as well and college player similarity to pros that could be helpful.

Neil Paine probably also has some stuff in the blog at basketball-reference.com on valuation.

[mtamada]

Be sure to properly cite the work of the people whose work you use. Note that there are new formats for doing citations of work found on the Web.

Why the importance of citations? If I'm the instructor, a student who looked into research already done by others rather than trying to re-invent the wheel, and then added his or her own insights, ideas, different variables, functional forms, etc. would be doing exactly what I'd want to see in an econometrics paper (or any type of research by anyone, for that matter).

A student who looked at research done by others, and simply copied what they did and presented it as his or her own work, is plagiarizing. And would be lucky to get an F for the paper (at some schools, an F for the course; and at a few schools, expulsion from the school).

There's a middle ground too: look at the work of others, add your own ideas, present the results. Sounds like the good situation above? Yes, but without the citations, the instructor doesn't know whether it's the good situation or the unfortunate second situation ... moreover uncited use of the work of others is, once again, plagiarism.

So citations are everything. Do look at what's been done before (in some fields, this would be called the literature review). Add your own ideas. And cite cite cite.

[Crow]

You could also search the internet for old John Hollinger Draft Rater articles about NCAA-NBA trends / specific player projections.

Here is some recovered discussion of them:

Author Message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Mon Jun 23, 2008 11:28 am Post subject: Hollinger's Draft Rater articles Reply with quote
John, I like the new third year PER projection scale.

I was wondering if you had any further observations on the indicator value of "program quality" (apart from your team strength adjustment of player stats due to competition for shots, rebounds, etc.).

Some GM's seem to stick to big name schools (top 20-30?) more than others and I wonder if that returns additional value on average, beyond what is indicated by your method. Portland is one tagged as going that way recently. Using the recruiting judgment of top coaches (with the most choice) as a future quality indicator or believing that they add value in terms of learning and locking players into the fundamentals makes sense but of course there will be guys who don't really take permanent advantage of it and guys who are better than the quality of their program origin.


I wonder if any additional defensive stat that includes shot defense ("defensive rating" or expected vs actual analysis of the box scores of player's main counterpart match-up) were added to the method if it would improve prediction of overall success (beyond just PER success).

I also wonder if players weighing over 250-260 in college or above a certain body fat % are giving off a warning signal that on average suggests likelihood of less than otherwise expected return.

Want to share anything further about your consideration of these or other issues?
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 245
Location: Pleasanton, CA

PostPosted: Mon Jun 23, 2008 2:04 pm Post subject: Reply with quote
I'm curious about Drew Gooden ranking as the 3rd best college prospect since 2002 -- after only Beasley and Chris Paul I think. Of course, any prediction system will have anomalies, but that one really stands out.

I know nothing about college basketball, but did that guy squander a lot of potential or what? Why does the system like him so much?
Back to top
View user's profile Send private message
kbomb



Joined: 09 Aug 2005
Posts: 9
Location: Gardena, CA

PostPosted: Mon Jun 23, 2008 3:50 pm Post subject: Reply with quote
is it just me or his year-3 projections look more like rookie year projections? i find it odd that none of the players studied has year-3 projection of PER over 20. to me, those numbers look closer to what they did in their rookie seasons.
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Mon Jun 23, 2008 4:30 pm Post subject: Reply with quote
Gooden got to almost 20 PER in his year 3, then lost 20% off his shot attempts per 36 minutes in his contract year. How much of that was him / warranted? I don't know the story step by step in detail but I know they got him re-signed at modest cost but then they got lower production thereafter too. How much was health related? In Chicago his numbers bounced positive from the lackluster previous 2 years. Wonder which level of production he brings next season.
Back to top
View user's profile Send private message
THWilson



Joined: 19 Jul 2005
Posts: 164
Location: phoenix

PostPosted: Mon Jun 23, 2008 6:13 pm Post subject: Reply with quote
I'd be interested in seeing if there's a correlation between the model residuals and draft position.
Back to top
View user's profile Send private message
Kevin Pelton
Site Admin


Joined: 30 Dec 2004
Posts: 979
Location: Seattle

PostPosted: Mon Jun 23, 2008 6:51 pm Post subject: Reply with quote
asimpkins wrote:
I know nothing about college basketball, but did that guy squander a lot of potential or what? Why does the system like him so much?

Pretty much any system you use will put Gooden that high, I suspect. His college numbers were off the charts.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
94by50



Joined: 01 Jan 2006
Posts: 499
Location: Phoenix

PostPosted: Mon Jun 23, 2008 10:55 pm Post subject: Reply with quote
Mountain wrote:
Gooden got to almost 20 PER in his year 3, then lost 20% off his shot attempts per 36 minutes in his contract year. How much of that was him / warranted? I don't know the story step by step in detail but I know they got him re-signed at modest cost but then they got lower production thereafter too. How much was health related? In Chicago his numbers bounced positive from the lackluster previous 2 years. Wonder which level of production he brings next season.

And Bulls fans everywhere wonder with you.
Back to top
View user's profile Send private message
tenkev



Joined: 31 Jul 2005
Posts: 20
Location: Memphis,TN

PostPosted: Tue Jun 24, 2008 9:04 am Post subject: One huge problem with Hollinger's Draft Rater Reply with quote
A huge problem with Hollinger's draft rater is that he uses PER as the dependent value and the flaws inherent in PER will be inherent in the Draft Rater. PER does a horrible job measuring defense and non-box score stuff like the value of a big man with range and of drawing double teams and setting picks. When is Hollinger going to get on the adj. +/- wagon? Its by far the most relevant measure to use in any regression analysis like this. It is more accurate with fewer biases and its high variation is not that much of a factor when it is used in this way.
Back to top
View user's profile Send private message Send e-mail AIM Address
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Tue Jun 24, 2008 11:02 am Post subject: Reply with quote
Quote:
When is Hollinger going to get on the adj. +/- wagon?


The availability of college basketball play-by-play data leaves much to be desired.
Back to top
View user's profile Send private message Visit poster's website
tenkev



Joined: 31 Jul 2005
Posts: 20
Location: Memphis,TN

PostPosted: Tue Jun 24, 2008 1:08 pm Post subject: Reply with quote
I wasn't talking about adj. +/- for college years. I was talking about the running the same regression off of third year adj +/- instead of third year PER.
Back to top
View user's profile Send private message Send e-mail AIM Address
John Hollinger



Joined: 14 Feb 2005
Posts: 175


PostPosted: Tue Jun 24, 2008 2:03 pm Post subject: Reply with quote
Adj +/- appears to have at least as many warts as PER, actually, most notably the "bad backup effect." PER is at least controlled more or less by the player in question, as opposed to what happens when he leaves the court.

As for Gooden, this is where good off-court scouting can come in. This rates pro potential, not what they'll actually become, and Gooden and Sweetney are examples of why sometimes "potential" never becomes "actual". Gooden can't remember plays and never improved, and Sweetney ate himself out of the league.

Finally, as to why no 20+ projections -- I'm looking at that. It may just be that it's more difficult to peg somebody as a surefire star than we think ... or it may be that my method is too tightly bound around the midpoint. The beauty of this is as we get more data with each passing year, the projections should automatically improve.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Tue Jun 24, 2008 2:10 pm Post subject: Reply with quote
Gooden in his year 3 rated the highest on adjusted +/- of any Cav, ahead of LeBron James. In year 4 Gooden ranked the worst Cav on the same measure.

Ferry takes charge and they shift from Silas to Brown, McInnis to Snow... and Gooden from burgeoning star to role-playing big?

In 04-05 Gooden-James worked well enough under Silas / McInnis (above James' average +/-) but under Brown/Snow Gooden became one of James' worst pairs.

How much blame should go to Gooden for weaker execution and how much is it due to different handling / perhaps role, less effective with and for him?

It is about team of course, not Gooden personally, and Cavs have had a fairly high level of success doing it their chosen way. But one step short of championship.


Gooden on his 3rd team by then (and now 4th) so the story is more complicated than just the numbers and I assume much of the buzz about him is fair. But he did improve from year 1 to 3 nicely on TS%, especially on DR%, assists, blocks, TOs. He can play well. And in short Bulls run was back to career average on TS%, above on both types of rebounding, at career high on assists, blocks and 2nd best on TOs. He can still play well in non-Brown / James - Cavs system...and least for a while.

It is true he has bounced around in his performance and that might continue too and it reflects on him and is not all about that system / role. But this year might help some with sorting out how much of the decline was "context" influenced. The system that was very good for James was apparently not as good for Gooden. Could things have been made to work better- for both and ultimately the team? More like under Silas (a big, perhaps both more sympathetic and more knowledgeable about how to use an inside big)? I don't know, just arriving to the issue. Those closer know how hard it was worked and everything involved. Not every player pairing or player=coach pairing or player-role assignment-team design works out.

Last edited by Mountain on Tue Jun 24, 2008 8:06 pm; edited 5 times in total
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Tue Jun 24, 2008 4:12 pm Post subject: Reply with quote
Cavs better than +2 on adjusted +/-:

04-05 4
05-06 4
06-07 3
07-08 2 James and the traded Gooden

Interesting trend.

And James after being #2 in league on this measure in 05-06, declined in 06-07 and again in 07-08.


Over the same time-period Kobe's Lakers went from 4, 3, to I think 1... and then to 5 this season. LeBron and the Cavs might benefit from a similar movement back to broader group of positive +/- contributors- by support players "picking it up" or "the system maxing their potential contribution as well" or some combination.


With the errors associated with adjusted +/- measurements these trends might not be as clear as they appear. But it still gives food for thought.


Number of +2 or better players on NBA champions last 4 years: 5, 6, 4 and 6 this season.

Last edited by Mountain on Tue Jun 24, 2008 7:26 pm; edited 2 times in total
Back to top
View user's profile Send private message
kevC



Joined: 28 Jun 2007
Posts: 17


PostPosted: Tue Jun 24, 2008 7:12 pm Post subject: Reply with quote
John, how high would have Carl Landry have rated on your new system. I think he was your biggest miss in last year's draft.
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Tue Jun 24, 2008 10:16 pm Post subject: Reply with quote
Number of +2 or better players on this year's other main contenders;

(from basketballvalue / heavily affected by playoffs)

Detroit 5
Orlando 3
Dallas 5
Houston 4
Phoenix 3
New Orleans 4
Utah 6
Denver 3
San Antonio 3

(Portland 5, but not Roy by this measure)

Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Wed Jun 25, 2008 12:23 pm Post subject: Reply with quote
Cavs most used lineup compared to such for other teams was one of the least used in the league.

5 of the top 7 used lineups had Gooden and the results were great, very good, good and 2 mild negatives.

The most used (but still lightly used at about 4 minutes a game) lineup included Gooden and was very good at +8.4 per 48 minutes.

Small sample unreliability concerns of course but not sure why you wouldn't try the good to great ones more.

Gibson - Gooden - Ilgauskas - James- Pavlovic does not seem like an odd or obscure lineup to examine. And yet it was apparently not used over 40 minutes in 06-07 and only 215 minutes in 07-08.

Coaches have their reasons for what they do and don't do but I remain puzzled by examples such as this one- a lineup that "works" so lightly used.

I see that it was just slightly better than team average defensive efficiency and well above average on offensive efficiency. Maybe it was not favored because it was not in keeping with the sought dominant Defense first type lineup. But net efficiency that lineup and another similar style offense better than defense Gooden lineup were the 2 best performing lineups among the top 10 used. How much should system / philosophy direct vs the numbers about actual lineup net results?
Ben Wallace may have fit the philosophy or the determined role better but didn't bring better results.

Hot lineup chasing can certainly be overdone in a game or a week or month but lineup analysis for a season in prep for second half of season or playoffs or future makes sense to me and how well it is done is a good subject for more research. Study of the consistency or inconsistency of coaching lineup selection and the results across the league in depth would help put specific cases in better context.


More I look at lineup data the more I see. The only negative lineups in Phoenix's top 30 most used are all Barbosa lineups. etc.
Back to top
View user's profile Send private message
Jacob S.



Joined: 24 Jun 2008
Posts: 2


PostPosted: Thu Jun 26, 2008 12:50 pm Post subject: Reply with quote
Just to throw my hat into the ring, I've also applied regression analysis in an attempt to project the sucess of college players to the pro game. Rather than PER, I used WinShares as the dependent variable. Unfortunately, due to time constraints in the data collection process, I wasn't able to include strength of schedule, pace, age, or height/weight. This was primarily a proof-of-principle project, and by next year I hope to explore the effect of all of these variables, and I intend to use statistical plus/minus rating as the dependent variable. Most of the obvious outliers in the current projections are players who are either overaged (Joey Dorsey, at 24 years of age) or products of a small school (Kyle Hines, of UNC-Greensboro).

In spite of the inherent flaws with my current approach, I did get some promising results. Breaking it down by Points Gaurds, Post Players, and Wing Players, the model arrived at r-squared figures of: .86 (PGs); .71 (Post Players); and .49 (SG/SF). I'm obviously not placing too much faith in the projection of wing players at this point.

Taking this approach, Beasley doesn't look like nearly the star that some other systems suggest he will be. For his career, he projects to average 13.06 Win Shares, so at his peak he would be expected to score somewhere around 20 win shares in a season. He's practically deadlocked with Richard Hendrix (13.57) and Kevin Love (12.59), while Ryan Anderson actually looks like the best legit PF (17.41). When I looked at Beasley's scoring and rebounding numbers I expected him to rate off the charts, so this all caught me off gaurd (I've wondered if an adjustment for team strength might shake things up a lot), but the problem with Beasley appears to be his defense and especially his passing (which is horrible). The top two centers are DeAndre Jordan (9.70) and Marreese Speights (9.55), with Brook Lopez projecting as a complete and total bust. Here is a listing of the top post players (Lopez has a negative projection):

21.02 Joey Dorsey, Memphis
17.41 Ryan Anderson, California
14.42 Kyle Hines, UNC-Greensboro
13.57 Richard Hendrix, Alabama
13.06 Michael Beasley, Kansas St
12.59 Kevin Love, UCLA
12.20 JJ Hickson, NC State
9.70 DeAndre Jordan, Texas A&M
9.55 Marreese Speights, Florida
9.11 Darnell Jackson, Kansas
8.61 DJ White, Indiana
8.12 Devon Hardin, California

Joey Dorsey isn't likely to be as good as this projection. He turns 25 years old in December, and by that point most NBA players are in their prime, with the bulk of their improvement already behind them. I originally hoped the collinearity between games played and age would avoid this issue, but the case of Dorsey shows that age simply has to be accounted for in any projection system.

I suspect that Kyle Hines (a natural PF who is 6'6" and weighs in at 230 pounds), is an outlier, and that adjustments for strength of schedule and height/weight would have dropped him a bit. Neverthless, he was impressive in college and played well at pre-draft camps, so he's probably worth an early-second round flyer.

The points gaurds rate like this, with the rating representing average win shares per season:

25.57 OJ Mayo, USC
23.99 Mario Chalmers, Kansas
15.35 Stefhon Hannah, Missouri
8.05 Derrick Rose, Memphis
7.24 Russell Robinson, Kansas
4.53 Jerryd Bayless, Arizona
0.86 Russell Westbrook, UCLA

Mario Chalmers comes out smelling like a rose, whereas Derrick Rose looks like he should be solid, but not outstanding. Stefhon Hannah looks like he could be a serious sleeper candidate, if he could just stay out of trouble. Outside of this top 6/7, everybody else in the draft actually projected to a negative average Win Share, which means that they aren't likely to stick around very long. Ty Lawson looks like a solid prospect for next year, as his projection would have been third best (14.70) had he stayed in the draft. Also of note, when I run OJ Mayo as a PG, his rating is off the charts (25.57 projected Win Shares, per season). So far, it doesn't appear that this is an effect where SGs fare better than PGs, but I haven't had time to thouroughly explore this possibility. If he plays anywhere near that level, then a backcourt of Mayo and Wade could be one of the all-time greats.

I'm throwing these preliminary numbers out here for those who are interested, and I'll post a bit more about my methods in a later post, but I want to again point out that this model does not currently include adjustments for pace, age, schedule strength, or team strength.
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Tue Jul 08, 2008 9:00 pm Post subject: Reply with quote
More on method later will be welcome.

I am curious if Chalmers, Hendrix and Anderson deliver.
Back to top
View user's profile Send private message
ErichDoerr



Joined: 06 Jul 2008
Posts: 15


PostPosted: Thu Jul 10, 2008 9:47 pm Post subject: Reply with quote
I finnd John's articles (Euros, Bigs, and Smalls) very intriguing.

I also find a lot of merit in viewing year-3 projections, though am not surprised by the lack of high PER forecasts. A 20+ PER projection would infer a lot of confidence on the player's coming greatness, and typically these prospects exhibit extreme youth and come with limited comparable statistical backgrounds. John's analysis will likely just require a greater sample base before projecting 20+ PER futures.

That being said, forecasting 82% of such a value is quite impressive, and I look forward to the improvements to come.

[Johndoe804]

Thanks for all the information. I'll be sure to appropriately cite any material I use in formulating my model, and I'll be sure to share what I come up with on this forum.

[Mike G]

Jacob S.
Joined: 24 Jun 2008
Posts: 2
Just to throw my hat into the ring, I've also applied regression analysis in an attempt to project the sucess of college players to the pro game.
... Beasley doesn't look like nearly the star that some other systems suggest he will be. For his career, he projects to average 13.06 Win Shares, so at his peak he would be expected to score somewhere around 20 win shares in a season. He's practically deadlocked with Richard Hendrix (13.57) and Kevin Love (12.59), while Ryan Anderson actually looks like the best legit PF (17.41). ... The top two centers are DeAndre Jordan (9.70) and Marreese Speights (9.55), with Brook Lopez projecting as a complete and total bust...
Mario Chalmers comes out smelling like a rose, whereas Derrick Rose looks like he should be solid, but not outstanding.

I assume these 'Win Shares' are from the old 3X formula.

[Bobbofitos]

I assume these 'Win Shares' are from the old 3X formula.

I assumed this too