College Basketball Rankings - Updated Daily
 |
| Are you suggesting that I, the president of Huxley College, go into a speakeasy without even giving me the address? |
My methodology page has more background, but the point of these rankings is to figure out what (and how) Vegas "thinks". I use the point spreads and over/unders to reverse engineer an implied ranking, called Generic Points Favored (or GPF). It's what you would expect a team to be favored by against an average team on a neutral court. By combining the point spread with the total, I can decompose team strength into its offensive and defensive components (oGPF and dGPF).
I first published these rankings for the 2011-2012 college basketball season, not long after launching this blog. At the time, I employed an Elo-style ranking system (see my original post for the details). Since then, I've hit upon a better way to create these rankings. As mentioned in this post on my MLB rankings, the new methodology weights recent games more favorably in a way that assumes that a team's ranking jiggles up or down randomly over time (a similar assumption is often used to model stock price movements). Bottom line, it's less kludge-y, less volatile, and, most importantly, more accurate in predicting point spreads for upcoming games.
Methodology
After working through rankings for the NFL, MLB, NBA, and NCAA Football over the past year, I've got a generalized methodology that works pretty well for any sport with betting action. It's only a matter of optimizing the parameters. Being a hopeless tinkerer, I did, however, make one minor change to the weights. Instead of using days elapsed in the weight function, I now use games elapsed. As a result, the rankings are less prone to the random jaggedness you see in my NBA rankings. When I get the time, I will probably revamp the NBA rankings to use games elapsed as well.
Parameters
- Home Court Advantage - Home court advantage is assumed to be worth 4 points, same as last year's rankings.
- Weights - Each game is weighted according to the following formula:
- weight = 1 / (0.5 + games elapsed), where the most recent game has a weight of 2, the one prior a weight of 2/3, etc.
- Game Outcome Adjustment - As I point out in my methodology page, gamblers are bayesians, and will readjust their thinking based on the outcome of each game. Game results are prone to randomness, so the market only partially corrects for an outcome that deviates from expectations. By optimizing the prediction error of future point spreads, I found that the market treats each game with 15% credibility, roughly similar to other sports. So, if a 3 point favorite wins a game by 9 points, the market would set the point spread for a hypothetical rematch at 4 points (3 points plus 15% of the 6 point "miss").
- Rest - Back to back games are relatively rare in college basketball, aside for conference tournament time, so I don't have any adjustments for rest like I do for the NBA.
The Ranking Table
For those of you that followed my recent rankings for other sports, the table format should look familiar. You can mouse over the column headings for explanations. Off to the right of the table, I compare my rankings to three external benchmarks: The AP Poll, the Coaches Poll, and Ken Pomeroy's rankings. As with last year, my rankings track pretty closely to Ken Pomeroy's, which is not surprising, as both rankings are intended to be predictive. The AP Poll and the Coaches Polls are more about rewarding season to date performance, rather than predicting future results (a distinction lost on many).
To keep the table manageable, I only show the top 64 teams (plus any team that is out of the top 64 but in the top 25 of either the Coaches or AP Poll). The sparklines for GPF, oGPF, and dGPF are all on a common scale, with the #1 rank at the top and the #64 rank at the bottom. Gaps in the sparklines indicate where the team dropped out of the top 64.
Leave a Comment