In this model, we simply consider the percentage of games that a team has played in previous weeks and either won or drawn. The Win-Lose call for a given game in the forecast week is based on the relative win percentages of the two teams playing. If the home team has a percentage greater than or equal to the visiting team, this model predicts the home team wins. If the home team has a win percentage less than the visiting team, then the mode predicts the visiting team will win. Note that this gives a slight bias to the home team.
The NFL 2014 season performance of this algorithm is shown below:
For the 2014 season, the call accuracy is 66% or correctly calling two thirds of all games. when evaluated over the 2012, 2013 and 2014 NFL seasons, the overall accuracy is 63% - which is noticably better than the 'Home Team' model.
The above model does not account for the home-team advantage pointed out previously. One possible way to take this factor into account would be by scaling the percent wins for each team by a home or away factor. Since the home team wins about ~57% of the time, and loses ~43% of the time, these can be used as scaling factors - not strictly kisher from a probability and stats perspective. Doing so improves the 3-year correct-call average from 63% to 64%, which is not a significant improvement.
Professional odds-makers have better predictive methods and algorithms than this one. This algorithm will NOT let you beat the odds in a consistent manner in Las Vegas-style gambling.
Next Page - Mean Points Model