Ever thought of developing your own ranking system for sports?
Over at bettingexpert, I wrote about the 5 qualities of elite computer rankings.
Four of these factors were relatively obvious to me when I developed the team ranking system for The Power Rank. However, I got pretty lucky with the last factor.
All 5 qualities are essential to the accuracy of the predictions that my rankings make. For example, the higher ranked team has won 66.8% of college football bowl games since 2005 (picked 177 of 265 games).
I don’t get into any technical details in the article. So don’t let the thoughts of solving linear equations or markov processes scare you from clicking on the link.
To check out the article, click here.
I’ve looked very carefully at the value of MOV for ranking, and in every test I’ve ever run, diminishing the value of blowouts (as you suggest in the linked article) reduces the value of the ranking for predicting future outcomes. Admittedly, I look primarily at college basketball, but I’ve also looked at college football and the same holds true. I’d like to see any evidence you have that reducing the MOV for blowouts improves performance.
Scott, thanks for the comment. I certainly believe you that diminishing returns might not work for every system. It depends on the details. I do know that my predictions would be way off if not for this feature.