It's been awhile since I posted on here after I stopped selling picks as a pro capper. I hope everyone is doing well and had a good football season. I'm in business school now and am running data regressions on college football data from the 2013 season. My goal is to develop a data model that will be able to predict what teams to target at the start of the football season.
I have started running regressions and have found out that things like penalty yds/game, 3rd down conversion %, turnover margin, and red zone percentages were not significant in predicting last season's ATS results.
This are just very preliminary results but points allowed, yards given up on defense, and the previous year's wins (negative correlation) were the strongest predictors. Basically, the higher the win total from the previous year led to worst ATS results the next year. Offensive stats didn't matter too much. Also, the current season's win percentage mattered too. Basically, it was hard for teams to consistently cover games but lose them outright. These factors were able to predict about 60% of the ATS results for the current season.
Essentially you want to target teams that didn't play well the previous year, have good defenses, and will win games outright. Each conference had about 2-3 teams that were good ATS winners, with most teams being around 50%.
Some other interesting notes were the following: once teams start giving up much more than about 25 yds per game, then their chances of being profitable as an ATS team are pretty slim to none, most good ATS teams scored an average of about 35 pts per game.
I'm still running regressions to try and find the variables that make up the remaining 40% of variance. I was thinking things like home attendance per game, career head coaching record of the coach last year, experience on the offense, defense, or combined team? What do you guys think? Any suggestions?
-Evan