NBA Rankings: All Star Breaks and Last Call before the Trade Deadline

You can recognize truth by its beauty and simplicity. When you get it right, it is obvious that it is right — at least if you have any experience — because usually what happens is that more comes out than goes in.
-Richard Feynman, The Character of Physical Law

I did something very different this year. Every previous year, I put all my numbers together before the start of the season and give you a win prediction for every team. I do this by trying to model and account for the eventualities and vagaries of the NBA season, coaches, players and rosters. I then reduced that complexity into a single number for each team. This year I did not do that.

28.3

This year, I built my model but then proceed to keep the uncertainty of the prediction and give you a range for wins. The point of this was to get a feel for how well I can predict that uncertainty and the value of doing this whole prediction exercise in the first place. Given that this is the last time prior to the deadline we get to do the rankings, I feel it’s also a perfect time to take a look a evaluating the model so far.

The key questions being: did I spend my time in the offseason productively? Am I approaching the truth or simply spinning my wheels?

Strap in because, I broke out the math for this one and made some enhancements.

Last time around, commenter T recommended that I use a Kalman Filter rather than a simple average of the season and the last ten games. ( Editors Note- A much more in-depth look at a similar method can be seen here care of Jirka Poropudas) This sounded simpler than it was but I did it anyway because it was a really good idea. The reason for the complexity came from the fact that this technique is sensitive to the ordering of the inputs and isn’t as sensitive as i’d like to variations. I fixed that by setting some controls to reset the gain on the input for large variations to the team value. I also built in a filter to protect against outliers ( Houston’s 45 point win in Utah on 1/28/2013 comes to mind) warping the results unnecessarily. These games will still be considered but a cap was placed against the actual maximum impact of a single game. What the result of all this work?

That’s a pretty but a little noisy. Would you like to see it in table form?

As currently configured, the model is producing an absolute average error of 9.5 points from the predicted margin to the actual game margin calculating based only on data available at game time. This compares to 9 points per game when I actually use total season point margin values for team. In layman’s terms, that’s pretty darn good.

I do need to take the time to run this in comparison with public markets in the near future (once I build a tool to accumulate the data).

Coming soon, quite possibly, to the internet :-)

We’ll come back to this, but let’s talk first about the actuals. As Always we start with everyone’s point margin adjusted for homecourt advantage and opponent.

The table has a running total by time period (season, last 30, last 25, last 20, last 15, last 10, last 5, and last 2) for the real point margin calculated for the period. The big change is that I added the Modified Kalman Filter estimate ( Point Margin Now) and sorted by it. This is meant to showcase the ebbs and flows of each team as the season goes along. It does however get more informative if I:

• Turn it into a win % using my own special formula.
• Color code for also rans (<44 wins), playoff teams (44-51 wins), contenders (52 -59 wins) and Champ Level teams (>60 wins).

San Antonio and Miami are starting to separate from the pack as clear finals favorites. The Thunder and Clipps are right there but a lot depends on which team shows up (OKC) and health (CP3).  The Hornets (as predicted by yours truly at the beginning of the season) have pretty darn good team once they have Davis available and you play a lot less Austin Rivers.

Their schedule is murder though.

The Hornets, Rockets, and Warriors have faced the roughest schedule (all out West for some reason). The Nuggets, Utah, Bulls, Pistons, Hawks and Pacers have faced the easiest schedule (all out East or at a mile high for some reason).

How’s it look for the rest of the season?

Again, I did simulate every game for the rest of the season to do that table. You may notice that I added what the delta is for schedule strength in win percentage before and after.

• The Hornets can’t catch a break and stay with the toughest schedule. Portland and the Kings also see a significant increase in difficulty.
• Warriors fans should not buy any playoff tickets.
• Denver and Utah continue to see a significant softening of their schedule. The Pacers, Hawks and Pistons get some help from the league office as well (and the Raptors in the Pistons case). I’d love it if it were possible to bet against Denver and the Hawks in Round 1 of the playoffs right now.

Let’s talk rankings.

The Rankings as of 2/14/13

The game data is courtesy of Basketball Reference.

The rankings are built by working out the following numbers:

• Point Margin per Game: (Pts scored by team) – (Pts scored by opponent) / games played
• Home court Point Margin per Game: Point Margin per game due to the schedule and homecourt advantage.
• Adjusted Point Margin per Game: (Point Margin per Game) – (Home court Point Margin per Game). Schedule independent point margin (neutral site at sea level)
• Adjusted Opponent Point Margin: The average Point Margin per Game of a teams opponents.
• Real Point Margin (RPM): (Point Margin per Game) – (Home court Point Margin per Game) + (Adjusted Opponent Point Margin). Expected Point Margin at a neutral site against perfectly average opposition.
• Point Margin Now:This is RPM estimated using the modified Kalman Filter. This is the Number I use to rank.
• Neutral Site Win % : A win projection using the Point margin Now and the relationship between point margin and win% (RPM/31 + .500 is a quick shorthand but not quite right, we gotta have some secrets)

As always, keep in mind that this is a guess (buyer beware) at the relative strengths of teams based on the data of the season to date with some weighing put in for more recent games. A more accurate projection would account for injuries and incorporate what we know of player historical performance. We will address this in a, say it with me, future post before the playoffs.

Let’s do some notes:

• Feel free to write in the Heat in pen to make the finals. They are clearly separated from the rest of the pack in the East. The Knicks would be the only second team that have a shot at them with the current roster and some health luck. Boston and Chicago could have a shot but not with the players suiting up as of today. The tradde deadline could play a role here.
• The West is insanely strong but the Spurs are in class by themselves. The three top teams would be sure things in any other year. I do think the Clippers are the best dark horse bet because, well, 40 minutes of Chris Paul. You know, they wanted to name a street after CP3, they couldn’t do it. Nobody Crosses CP3 he crosses you

Let’s show you the full season simulation now. One thousand sims for every game using the New Point Margin Now Model.

That’s a range breakdown by games won and by team. You’ll note that the Spurs are 5% to win 67 or more games. Crazy. Let’s look at it in table form. I’ve included the current projections sorted as minimum, average and maximum projected wins. The table also shows the numbers from the preseason projection and whether that number was low, high or in range based on the current status of the league. A high means the current high is below the average for the preseason, a low means the current low is above the average for the preseason. I’ve also included the net playoff odds.

26 of the 30 projections are in line with the preseason numbers. For the lows, we have Brooklyn (Brook Lopez as a legit all star), Detroit (Drummond even though his season is done),  and for the highs we have injury mavens the Twolves and Lakers. The 86.75 hit rate however does indicate that the preseason prediction was time well spent.

Let’s talk playoff odds in detail.

The top seven seeds (barring trade, catastrophic injury or unexpected comeback) are pretty much set. The dram in both conferences is in the eight seed. I do think the East race is tighter than shown because:

• The Bucks want to make a trade and there’s no guarantee it’ll be good.
• Bargnani is poised to make a comeback in Toronto (sorry Raptors fans).
• Bynum might actually play this season.
• Calderon is a good player for the Pistons.

The West seems to be much more wide open with the Warriors crashing back to reality and the Mavs, Hornets and Lakers all lurking. The magic number looks to be 45 wins. The key factors here are:

• What if anything can the Warriors get from Bogut?
• Can Dirk get back to a decent form for the Mavs?
• Will Gordon force Austin’s playing time to dwindle?
• Will D12 get better over the second half?
• Will Portland collapse against their killer sched?
• Can Minnesota field a somewhat healthy team?

I tend to think that Philly and the Lakers are probably still the best bets for the honor of getting creamed in round 1.

As I’ve said before: before the season, I picked San Antonio over Miami in the finals. I see no reason to change my mind.

-Arturo

P.S. Again as a bonus, Here’s a table presented without explanation or context: