As mentioned on my previous page, OAF was developed in the summer of 2002. Since I was attempting to design an algorithm that would objectively approximate how the polls evaluated teams at the end of a season, I did not think of using this "system" to predict outcomes until December of 2002. Since post season bowl games are typically held at a neutral site, concerns related to considering home field advantage could be ignored which is imperative since OAF is a ranking (and not a rating type of) system. So, I studied how the seven computer systems used in the BCS formula did predicting those 28 bowl games as well as the "Las Vegas line", the majority team selected by the "BCS seven", and OAF. As it turns out, OAF performed really well that year compared to the others (even better than I would have guessed as described later on) and so I continued this "experiment" over the next two years.

I learned about Kenneth Massey's comparison page (listed on the previous web page) and downloaded the rankings of all those systems for 2002 to find the BCS seven "full rankings". I have now exported the rankings of all those systems into an Excel spreadsheet and have analyzed how all of those systems have done since 2002. (Summaries for those years are contained below, and full pages, with rankings for each team in a bowl game, for each and every system, along with which bowl games which system correctly/incorrectly predicted, are on separate pages. These pages are quite large, and probably will take some time to download if you are really interested in seeing them).

As the table below summarizes, there are quite a few systems on that comparison page, and 58 of them are included in each of the four years, so how well those systems have performed is summarized at the bottom of this page in descending order of "prediction success". In 2002, only 2 of the 78 systems had 18 right (OAF was one of them!), and that year is strange in the sense that 80% of the systems correctly chose the right team in 14 of the 28 games, but in 12 of the other 14, the correct team was selected in less than 20% of the games. (There were 9 games were all systems were unanimous in their selection, and 6 games went according to plan, and the other 3 must be considered as "upsets". There were 26 systems with 14 correct predictions, and a different 26 systems with 15 right, which is why the median came out as 14.5. The number in the parentheses are the counts for those totals.)

OAF correctly predicted 18 out of 28 games in each of the first three years it existed, which as it turns out, was outstanding in 2002, but only average in 2003 since that was the median number of correct predictions that year. In 2004, there were only 5 systems with more than 18 correct predictions, and 6 others with exactly 18 right. In 2005 OAF underperformed, as it was the first year it didn't achieve 18 correct (only 15-13 ), but it can back strong in 2006 with 22 right out of 32 games, leaving it as the 4th largest overall total of correct predictions of the 45 systems that have appeared on Massey's page from 2002-2006. 2007 was a little lackluster, with only 19 wins, but OAF is still tied for 6th overall over the past 6 years.

Here are the summaries for the previous years. (All the large data files that were used to construct the tables below can be found at: 2002, 2003, 2004, 2005, 2006, 2007.)

(KEL and RAN only appear on Massey's rating page in 2002.)