I'm still processing West Virginia's demolition of Oklahoma.  I know most people think this season was one of the best in history because of all the astounding upsets, but my personal opinion is ... this season sucked.  I hate College Football.

Last season AccuScore's point spread prediction accuracy was 59% (386-268).  That's pretty impressive.  While college football totals have been good all season (56% for the year), I am pissed about this season's point spread performance.  This season was humming along just fine untilWeek 5 when the first huge upset occurred - Colorado beats Oklahoma.

Point Spread Record Weeks 1 - 4:       99-81 (55.0%)

Point Spread Record Weeks 5+:          236-241 (49.5%)

As someone focused on using objective statistics to forecast games, it really pisses me off that so many teams generated statistics one week that seemed to have no relevance or predictive value for future weeks.

Take West Virginia and Oklahoma.  I know Pat White was injured, but how does the team run for 104 yards against Pittsburgh to lose 13-9 and then rush for 349 yards against Oklahoma? 

This is an Oklahoma team that just held Missouri to 98 yards in their previous game.... That's a Missouri team that just ran for 323 yards against Arkansas and won 38-7.... That's an Arkansas team that just beat LSU, the team playing for a National Championship. 

Using the transitive property, Pittsburgh just beat LSU?!!! 

To link these massively contradictory results you only have to use each team's past 2 games.  There were so many cases where a seasons' worth of statistics provided meaningless value to predicting how teams would play in the final game.

Take the WAC.  Nevada averaged 36 pts per game.  They score 0 against New Mexico.  New Mexico gave up 34 points to New Mexico State, another WAC team that Nevada racked scored 40 pts against!

No one expected Hawaii to score 46 pts (season average) against Georgia, but to score just 10 (7 in total garbage time), is abysmal.  AccuScore simulations had Nevada and Hawaii scoring 33-40% fewer points than average in their bowl games, but these predictions were still way off.

A few things you knew about the Michigan-Florida game was Michigan struggled to stop the spread offense, Mike Hart never fumbles, Tebow was the star QB, not Chad Henne and turnover margin is the best indicator of who will win or lose.   So what happens?  Michigan handles the spread just fine, Henne has a career day, Mike Hart fumbles twice on the goal line and Florida loses despite a 4-0 edge in turnovers!

The list of freak occurrences goes on and on. 

In the end, AccuScore rode quality over-under performance to remain 30% profitable in college football, but the decline in point spread accuracy was clearly tied to the fact that this season was once in a life-time.  If point spread accuracy declined due to consistent slight inaccuracy then I would expect off-season back-testing of simulation accuracy would help create a "better" model going forward. However, with all the whacked out results and lack of consistent data, I highly doubt there will be any enhancements that would show improved accuracy over the past 3 seasons.

I think this season was a lot like the 2006 College Basketball season where George Mason made the Final 4 and there were a ton of upsets in the NCAA tournament.  Many people thought this was a sign of things to come, but in 2007, the favorites led by the most talented teams (ex. Florida, Ohio State, Georgetown, UCLA) dominated the tournament. 

I expect the 2008 College Football Season to "behave" properly and more predictably and I will once again go back to loving college football.