Saturday, November 24, 2012

NBA Team Rankings Now Available

NBA team rankings are now available for the 2012 season.  For those of you unfamiliar with my ranking methodology, the goal is to reverse engineer an implied power ranking from the Vegas point spreads.  I'm trying to figure out who the betting market "thinks" are the best teams in the league.  In addition, by combining the point spread with the betting over/under, I can decompose team strength into its offensive and defensive components.  My Methodology page has a simple example of how the rankings work.  You can also refer to my series of posts at Advanced NFL Stats Community, where I first created these for the NFL.

Here is a direct link to the ranking table.  As with my NFL, NCAAF, and MLB rankings, these update every morning with the latest game results and betting information.  My source is Sportsdatabase.com.


Ranking Table Overview

Here is a description of the fields (you can also mouse over the column headings):

  • LstWk - The team's ranking as of a week ago, with a corresponding sparkline that shows the day to day movement (mouse over for the actual numbers).
  • GPF - Stands for "Generic Points Favored".  It is what you would expect the team to be favored by against a league average team at a neutral site.
  • oGPF - Stands for "Offensive Generic Points Favored".  The component of a team's GPF that is attributable to offense (and pace).
  • dGPF - Stands for "Defensive Generic Points Favored".  The component of a team's GPF attributable to defense (and pace).
  • GOU - Stands for "Generic Over/Under".  It is what you would expect the betting over/under to be set at when playing a league average team.
  • W-L - Team win-loss record (it is interesting to see where the betting market diverges from a simple win/loss ranking)

Some Initial Observations

  • The Lakers - Note the rapid rise in their Generic Over/Under, corresponding to the hiring of Mike D'Antoni, a coach known for fast-paced, high scoring offenses.  So far though, the Lakers' overall GPF ranking has not moved much, meaning that the market isn't convinced yet that D'Antoni's style will bring wins as well as scoring.
  • The Bobcats - The market is playing "wait and see" with the Bobcats so far.  It's going to take more than a 6-5 start to erase the memory of last season's historically bad 7-59 record.

Methodology

The methodology went through some slight tweaks this year, mainly in how I weight prior games.  Recent games are given more weight since I am trying to get the most up-to-date estimate of what the market thinks.  The weight I use is as follows:

weight = 1 / (2 + days elapsed)

For example, today's games would be weighted at 1/2, yesterday's at 1/3, etc.  The factor of 2 in the denominator was chosen so as to minimize the prediction error of future point spreads (backtested against prior NBA seasons).

Home field advantage is assumed to be worth 3.25 points.  Teams playing back to back are assumed to be penalized 1.25 points in the point spread.

The market is assumed to treat each game outcome with 20% credibility.  For example, if a team was favored by three points and instead won by 8 points, the revised point spread for a hypothetical rematch would be 4 points ( = 3 points + 0.20*(8-3) ).  This revised spread is what is actually fed into my model.  I add this adjustment because it improves the prediction error of future point spreads.

Next Steps

Now that the ranking table is up and running, I hope to relaunch The Ticker and Today's Games features soon. 




3 comments:

  1. Hi Mike,

    If the market treats each game outcome with 20% credibility, does this mean that if a 3-pt dog won the game by 8 that the revised point spread would be +1 (0.20 * 8+3 = 2.2)?

    Thanks.

    ReplyDelete
    Replies
    1. The revised point spread in that case would be:

      80% x 3 + 20% x (-8) = 0.8

      Delete
    2. Thank you for getting back to me with the exact math involved.

      Do you think there is some recency bias baked into these formulas? I ask because when we consider that a 3-pt favorite in basketball will win the game SU about 57% of the time and a 1-pt favorite will win the game SU about 52% of the time, is one data point or result really worth a 5% shift in win probability?

      Also, does the zig-zag theory in the playoffs run counter to these findings in the sense that if a favorite loses straight up, investors in the marketplace will often play back on the favorite the next game. As a result of this practice, the favorite’s price point will often be the exact same as the previous game or perhaps another point higher (providing at same venue).

      Thank you.

      Delete