So far for 2013-2014 my model using last year stats went 1-4. Honestly doesn't concern me one bit. I noticed the difference between the previous year's data and what actually took place was substantial. That is why I believe you can't begin betting until Week 5...not enough stats to go on. With that being said, I decided to use the finalized stats... plug them back into the model and see what it would have predicted. I DONT use points for or against in order to determine my over/under number so it really doesn't matter if they score 60 or 40... With that being said...I have created a model based on cumulative statistics year to date... I like my relationships within the model that I have established. But I have constructed my win percentage based on cumulative data from each team competing against cumulative data from the other team.... It occurred to me that I should be looking at a week as a singular event. Than I can optimize my model around actual statistics on a certain week and compare the likelihood of a team using their cumulative data... to hit those actual statistics. I am sure I am doing an awful job of explaining it... but should I optimize a system based on data using actual statistics at the end of a weekend or optimize a system based on cumulative data vs. cumulative data.
I also wait until week 5. I use rolling 4 week averages and 3 week averages in my model. 4 week average is smoother with more variation in 3 week data. 3 week data though can help identify a team breaking out of a trend though there can be mis-direction. I don't believe in cumulative data since a team doesn't play the same way in week 1 as it does say in week 15.