The Milwaukee Bucks won 46 games in 2009-10. This result was a surprise to the pundits, who expected the Bucks to be among the worst in the NBA. So how is it the NBA came to “Fear the Deer”?
Ty Willihnganz at Courtside Analyst summarized Milwaukee’s 2009-10 season last April with the following comment: Berri and I and BucksNation were all right
Last night the Milwaukee Bucks clinched a playoff spot.
On October 29, 2009, a certain author of a certain book you ought to purchase and read sent me the following email, Subject lined “Bucks forecast?”:
Hi Ty,
What is your forecast for the Bucks right now? Still think they are going to be good? The consensus is they are very bad, but I don’t see that in the numbers.
I didn’t see it in the numbers either. And I said so at the time. And the readership here was with me, and now we have been vindicated.
Skiles rules, the Bucks rule, and BucksNation rules. And this should teach us all something about the reliability or credibility of the national press corp.
In the summer of 2009, the pundits argued the Bucks would be one of the worst teams in the NBA. Ty and I thought otherwise. And when you look at the numbers from 2009-10 – in the table below — you can see why Ty and I did not think the Bucks were going to be among the NBA’s worst last year:
A simple look at the past numbers suggested that the Bucks would win about 35 games last year. And this forecast is artificially low because the only numbers we had on Ersan Ilyasova were from 2006-07 (his only prior NBA season when he was quite bad).
Why did the pundits think the Bucks were going to be extremely bad? Once again the problem is with the perception of scoring. The Bucks only won 34 games in 2008-09. And then Richard Jefferson (19.6 points per game) and Charlie Villanueva (16.2 points per game) departed. Furthermore, Michael Redd (21.2 points per game) was injured and there was some question if he could play the entire season (he ultimately only played in 18 games). With all these scorers departing (or not contributing), the pundits thought a 34 win team was about to get much worse.
Ultimately the leading scorer on the Buck was John Salmons (19.9 points per game). But Salmons wasn’t there at the beginning of the season. If you look at the players who were there from the beginning, the leading scorer was Andre Bogut. And Bogut only averaged 15.9 points per game. So as the pundits expected, the Bucks did not have a dominant scorer for much of the season. Still, this team was not horrible.
To understand why, we need to look at all the stats (via Wins Produced). And when we take this step it was clear the Bucks were not going to be horrible in 2009-10. Bogut, Carlos Delfino, Luke Ridnour, and Luc Mbah a Moute had all been productive players in 2008-09. And since NBA players (relative to what we see in the NFL and MLB) are quite consistent across time, it was reasonable to expect these players to produce win in 2009-10.
That is what happened. This quartet produced 29.4 wins last year. So even if we only look at these four players, we would have expected the Bucks to be better than the 14th best team in the Eastern Conference (where ESPN.com placed the Bucks) or the 28th best team in the NBA (where Chris Mannix at SI.com ranked the Bucks).
What about this next year? The pundits have seen the Bucks were a playoff team last year. And the Bucks have not lost any major scorers. So the pundits seem to think the Bucks will be about what they were last year.
What if we turn to Wins Produced? As the Bucks story last year demonstrated, forecasting the NBA is easy. Right?
Well, before we get too smug, let’s look at some forecasts.
Let’s start by noting that this year the pundits and Wins Produced seem to agree with respect to the Bucks. Ty Willihnganz projected the Bucks twice (see HERE and HERE) and each time concluded the Bucks can expect to win about 46 games in 2010-11.
Ian Levy – at Hickory High (the latest member of the Wages of Wins Network — and past contributor in this forum) — utilized Wins Produced to forecast every team in the NBA.
As you can see, Ian concludes the Bucks can expect to win 45 games next year (Ian also has an Expected Point metric that looks quite interesting – and I hope to talk about this in the future).
Ian is not the only one to use Wins Produced to forecast this next season. Arturo Galletti is also busy forecasting. And his latest (he is offering various versions) says the Bucks can expect to win 39 games. So Arturo expects the Bucks to decline a bit.
Six or seven wins, though, are not that big of a difference. An injury here or there can easily cause a team to decline by this amount. Likewise, if Brandon Jennings improves (young players do this) then the Bucks can be better than Arturo expects.
One should note that Ty, Ian, and Arturo are not the only people who will be forecasting this next season. And what we see from just these three forecasts is that even if people mostly agree on how basketball should be analyzed, forecasts will still differ somewhat.
Maybe it would be better to do something similar to what Chad Ford does for the NBA Draft at ESPN.com. Rather than forecast exact draft position, Ford focuses on Tiers. In other words, some players are Tier 1, others are Tier 2, etc.. One would expect the players in the first Tier to be taken before the second Tier, but within Tiers it is hard to forecast.
Looking back at Ian’s forecast, one can argue the NBA breaks down into the following Tiers:
Tier 1 (championship contender – 55 or more wins): Miami, Portland, LA Lakers, and Chicago
Tier 2 (homecourt playoff contender – or 45 to 54 wins): Orlando, Boston, Milwaukee, San Antonio, Dallas, Golden State, Oklahoma City, Phoenix, and New Orleans
Tier 3 (possible playoff contender – or 35 to 44 wins): Atlanta, Cleveland, Sacramento, Denver, Utah, Memphis, and Houston
Tier 4 (top lottery team – or 25 to 34 wins): New York (could be playoff team in weak East), Charlotte, Indiana, New Jersey, Minnesota, Detroit, and Toronto
Tier 5 (bottom lottery team – or less than 25 wins): Philadelphia, LA Clippers, and Washington
Again, this is based on Ian’s foreacasts. If we use Arturo’s forecast, we see…
Tier 1: Miami, Portland, LA Lakers, San Antonio, and Boston
Tier 2: Chicago, Orlando, Utah, Atlanta, Dallas, Golden State, New Orleans, and Denver
Tier 3: Oklahoma City, Phoenix, Sacramento, New Jersey, Milwaukee, Houston, Memphis, Cleveland, Charlotte
Tier 4: New York, Detroit, Toronto
Tier 5: Minnesota, Indiana, LA Clippers, Philadelphia, and Washington
Okay, there are some similarities. But there are clear differences also. And again, Arturo and Ian are using the same basic idea to forecast the teams (so maybe the Tier idea doesn’t really work).
All of this illustrates something that I think needs to be emphasized. Even if we agreed on how to evaluate the past, projecting the future is still not going to be something we all agree upon.
And it also illustrates how I think that people looking to “test” Wins Produced might need to think of other tests. For example, one could consider…
1. Is there a clear theory behind the model (this is not the same as … does the model produce results consistent with your prior beliefs)?
2. How well does the model explain what it purports to explain?
3. How consistent is the model across time?
And these are just some of the issues to consider. For a semi-complete guide one could consider something I wrote about three years ago.
– DJ
arturogalletti
October 18, 2010
DJ,
Great, great piece (and super long!). You’re totally right that this isn’t about testing the metric (it is what it is an what it is is good :-)).
For me at least, it’s about building an effective model using sound principles. I do understand that there are multiple variables and that the likelyhood of getting it right (i.e. sound in theory, significant, highly correlated, robust) on the first go-around is,umm, let’s call it remote. For me, I’m very interested to see where the opportunities for improvement lie and what each particular model (Ty’s, Ian’s, mine, everyone else’s) is good and bad at.
ilevy
October 18, 2010
Agreed! This is the first year I have done this and I know the true test for me comes at the end of this season in evaluating strengths and weaknesses to make adjustments for next year. Great post, and not just because it mentions me!
entityabyss
October 18, 2010
I got a question on levy’s prediction (I assume that arturo’s is not done yet). With all the teams that won 50 games last year, why are so many teams falling off. Why are teams with new additions winning less games (la lakers)? Is the dropoff in the league that big this year?
Also, I got golden state for more than 50 games. It’s just too hard to see it not happening. Taking all their players, adjusting for minutes, and adding their WP from last year got them close to 60. That didn’t consider that they added someone (I forget who) who adds 5 wins and the expected from their sophmores. They should be good for at least 53. I’m not saying levy’s prediction is wrong or that my thought makes it wrong. Just giving my opinion. Is there reason to believe golden state won’t win 50 games?
ilevy
October 18, 2010
I didn’t do my individual players adjustments based on personal opinion, instead using the percent change in individual win% as projected by Basketball-Prospectus. (I also used their minute projections as a guide.) So I can’t speak to the rationale’s for every decline. In the case of the Laker’s the new additions (Matt Barnes and Steve Blake) project to big producers of Wins. The decline comes from slight individual declines for Pau Gasol, Lamar Odom and Kobe Bryant, likely age related. In addition Derek Fisher hs declined sharply but is still likely to see way more minutes than Blake. Finally Bynum’s injusry situation makes in unlikely that he will play as many minutes as last season.
As far as Golden State my assessment, which has them at 49.6 Wins, rounded down to 49, I think the major issue will be minute distribution. My projection has them with only one above average producer in the backcourt, Stephen Curry (WP48 0.163). They also have Monta Ellis, projected to play just over 2800 minutes with a WP48 of only 0.035. The bulk of their talent is in the frontcourt. By my projection they 6 above average producers in the frontcourt (David Lee, Andris Biedrins, Brandan Wright, Louis Amundson, Dorell Wright, Reggie Williams), to go along with 5 below average producers who will likely see minutes as well (Chris Hunter, Dan Gadzuric, Rodney Carney, Vlad Radmanovic, Ekpe Udoh). Keith Smart appears to be committed to a more traditional lineup and rotation, but regardless some of those frontcourt players will be squeezed for minutes and have their production underutilized. In the grand scheme of things it’s a good problem to have, but Louis Amundson with a WP48 of 0.182 won’t have a huge effect on the team’s bottom line win total if he only plays 700-800 minutes. If the team could turn one of those frontcourt assets into an above average producer to come off the bench at one of the guard positions it could really boost their potential this season.
I hope that helps, but please let me know if it just raises more questions!
kevin
October 18, 2010
I think Ian’s boston forecast is too low. They were a better team than they seemed because Pierce and Garnett were limping through the season and their rebounding numbers were bad. Garnett, especially, seems rejuvenated and the two O’Neals should help a lot on the boards.
entityabyss
October 18, 2010
Oh rodney carney and vlad radmanovich will take minutes. I get it now. I wanna see that win% thing, because using the espn fantasy stat projections, it didn’t seem like there would be a problem with minute allocations. Only 4 50+ win teams in the west from about 7-9 last year seems hard to grasp and miami as the only 60+ win team. I don’t necessarily disagree with this, but I didn’t see it coming. I get it though. I get why declines are expected.
Joe
October 18, 2010
kevin,
KG isn’t getting any younger. Maybe he will be better, but he is approaching the age where injuries seem to be constant.
some dude
October 18, 2010
The Bucks were 24-28 before the Salmons trade, which was unforseen.
So what explains this? Salmons only added 2.4 wins to the team according to WP, so how did they go from a bad team in the East to a good team in the east post-salmons trade?
I don’t think anyone thought Jennings would be where he was (and defensively he’s pretty solid) nor did anyone in their right mind expect Bogut to go from a mediocre big man shotblocker to a great shotblocker randomly. I doubt that’s common (Bogut is quite underrated even before the blocks, of course). And Ridnour had a career year at a late age. And like you said Iylasova was a mystery and reversed fortune.
Bucks look like a combination of too many unknowns + mis-season trade + career seasons from more than 1 player. Hard to take credit one way or the other.
however, I’d love to know why the Bucks became good after Salmons was traded to them despite his .13WP48.
arturogalletti
October 18, 2010
sd,
To be fair, the Bucks were on their way to an around 35 win season last year. Everything else was Ersan,Bogut and Salmons not sucking when he came from the Bulls (.069 WP48 for chicago, .128 for the Bucks).
The reason my model for this year doesn’t like them is it assumes a regression for Salmons, injury for Bogut and CDR actually getting on the court (which is of the bad).
robbieomalley
October 18, 2010
I think I’ve got my predictions all set and ready. But the more I go into them the less I feel like what I’m doing has any chance of being accurate. Worth a try though!
some dude
October 18, 2010
arturo, yeah i get that. But I don’t think anyone predicted they’d even be on a 35 win pace for the reasons they did. And in fairness, they weren’t predicted to win 35 pre-salmons. It’s was more like 32. 30-31 is we figure in Stackhouse signing after the season began. And then there’s the 4 wins from Jennings. Was that predicted? Now we’re down to 26 wins according to the model in pre-season not including Jennings.
Anyone predict Bogut to nearly triple his blocks? Or Ilyasova to go from awful to good? Or Ridnour to have a career year?
My point is that it seems like according to WP it’s more like a mid 20 win team. Better than some of the dire predictions, but also not the 35 claimed above an nowhere near the 46 they won (based on pre-season). What were the projections last year?
Also, I’d love my question to get answered. The Bucks went from a 24-28 team to 22-8 team in the Warrick for Salmons swap. What explains this when Salmons was only worth 3 wins? If you analyzed the Bucks WP at the time of the trade, would anyone have predicted that big change because of Salmons?
Chicago Tim
October 19, 2010
Hi Professor. Arturo has suggested another possible measure of a statistical model. How much money can we make betting on it? (Theoretically in states where it isn’t legal, of course.)
http://arturogalletti.wordpress.com/2010/10/18/nba-team-win-predictions-as-of-now-vs-the-lines/
I for one have made a note to come back to Arturo’s article at the end of the year and find out how much money one could have made. Theoretically.
What’s interesting about this method is that Arturo singles out the seemingly-sure bets that defy conventional wisdom. And he puts his money where his model is. Theoretically.
Bill Gish
October 19, 2010
If it were possible to reliably predict future wins based on past performance then gambling on horse races wouldn’t exist. Presumably, once the secret was out only the rubes too ignorant to apply the forecasting model would ever place losing bets. Racetracks would go broke making good on all those winning bets that had to pay the legal minimum of $2.10.
Tracks don’t go broke because it happens so rarely but this actually happens as explained in this quote:
Bill Gish
October 19, 2010
If it were possible to reliably predict future wins based on past performance then gambling on horse races wouldn’t exist. Presumably, once the secret was out only the rubes too ignorant to apply the forecasting model would ever place losing bets. Racetracks would go broke making good on all those winning bets that had to pay the legal minimum of $2.10.
Tracks don’t go broke but this actually happens as explained in this quote:
The Heat may generate a few negative pool hedge bets this year.
But I have a question. How did the models in question handle known unknowns ( to borrow Donald Rumsfelt’s expression)?
Blake Griffin’s easily imagined wins produced could surely raise the Clippers projection of just 22 wins
Chicago Tim
October 19, 2010
Bill Gish —
I know someone who claims to make his living betting on horses. I have no idea whether he is for real. There are lots of rubes, though. The people taking the bets don’t go broke as long as there are bets on both sides of the line. Don’t you think a statistically-savvy bettor could have made money on baseball in the hundred years before Bill James, Sabermetrics, and Moneyball?
John Giagnorio
October 19, 2010
It was possible to make money betting on baseball as recently as the early 90’s, actually, and I would imagine for a long time before then. The guy I met who claimed to have done it talked mostly about using DIPS. One of the very prominent sabermetrics writers made a ton of money betting baseball games.
Bill Gish
October 19, 2010
Let me amend my earlier statement. I added the bold face print.
“If it were possible to reliably predict future wins based on past performance then gambling on horse races wouldn’t exist. Presumably, once the secret was out and widely disseminated only the rubes too ignorant to apply the forecasting model would ever place losing bets.”
It was exactly my point that a bettor employing an accurate system of predicting wins would make lots of money until everyone else caught on.
So of course I am not surprised to hear that an early adopter of a system like sabermetrics made a ton of money just like early adopters of card counting in Vegas mad a lot.
Maybe there will always be enough rubes even among NBA GMs that followers of WP as a statistical model will retain an edge.
But I would really rather go back to my question of how Ian and Arturo and Dr Berri handle “known unknowns” like Blake Griffin’s pretty predictable boost to the Clippers?
arturogalletti
October 19, 2010
Bill,
My Clippers projection has Blake Griffin at 2447 mp (prediction for no.1 pick in the draft) and projected at .191 WP48 (+/- .065) so about 10 wins +/- 3 wins. The rest of that team just doesn’t project as very good.
ilevy
October 19, 2010
Bill,
My Clippers projection has Griffin at WP 0.150. With 2090 minutes played he would produce 6.53 Wins. I cheated and relied heavily on the 2010-2011 Basketball Prospectus for help with my predictions. I don’t follow each team enough to make accurate predictions on what their rotations will look like so I borrowed the minute predictions from Basketball Prospectus. For rookies I started with what I thought was the average WP48 for rookies, 0.050. I have since learned that it’s actually 0.042 which I will have to account for next year. I then multiplied this by the Wins Above Replacement Player projected by Basketball Prospectus. In Griffin’s case, 1.5.
EntityAbyss, the win% numbers are part of the 2010-2011 Basketbal Prospectus. You can go to BasketballProspectus.com, and follow the links to purchase a PDF copy (only $10). The win% predictions are listed for each player.
Somedude, my take on the Bucks improvement after adding Salmons was the way he increased the scoring efficiency of those around him. I would have to look at some more numbers to see if this really works out. My tentative guess is that he was a more efficienct scorer than most of the rest of the roster. By taking shots away from less efficient scorers he would have increased their individual WP48. So although he only added a few wins on his own, his scoring efficiency relieved scoring responsibilities from less efficient players thereby accounting for the uptick in their Win%.
some dude
October 19, 2010
ilevy, if that’s true, then that means WP is missing something big in the analysis that undervalues a player, such as Salmons, doesn’t it?
ilevy
October 20, 2010
This certainly sounds like a cop-out, but I feel like I don’t know enough about the stat to answer that question in any complete way. My instinct is that looking at WP48 numbers for a single season misses some of the effects that a player has on his teammates WP48. The WP48 numbers for Cleveland’s supporting cast make them look capable of competing for an 8th playoff spot this season. I have a feeling that may not be the case. I think it’s likely that some of those players will see their WP48 numbers fall off slightly as they may have been falsely inflated by playing next to LeBron. I think it’s possible that a similar thing may occur with Amare Stoudemire in New York. Scoring efficiency is a large component of Wins Produced and my instinct tells me that his scoring efficiency will suffer this season from not playing next to Steve Nash.
These are certainly extreme cases, as Nash and LeBron are two players who have a monumental impact on their teammates. This may be something that could be accounted for with a multi-year study of WP48 numbers and how they change with different combinations of teammates. Again, I don’t feel like I have the knowledge to answer this in any satisfying way. It might be a great question to send along to Dr. Berri.
GexRolaannele
November 23, 2010
That is another intriguing and comprehensive post , even if this is perhaps too advanced for those of us inexperienced in forex.
I would be glad to see a post on the basics everyone needs to keep in mind .