The problem with using only one model in football predictions

You might wonder why UCF is first in my public college football rankings after four weeks of the 2017 season. That makes no sense whatsoever.

Let’s break it down. First, UCF has played only two games this season due to the hurricanes in Florida. They stomped Florida International 61-17, and then they beat Maryland on the road 38-10, a game in which they were a 4 point underdog.

My team ranking algorithm takes margin of victory and adjusts for strength of schedule. UCF not only has a huge margin of victory, but they have a key win over Maryland.

Maryland makes UCF look good because they beat Texas 51-41 in their opener. The team rankings don’t know that Maryland QB Tyrell Pigrome got hurt in the Texas game and won’t play the remainder of the season.

Texas went to USC and almost pulled off the upset in week 3. This boosts the stature of Texas, which gives Maryland a bump and pushes UCF up higher in my team rankings.

USC destroyed a good Stanford team, which pushes USC up in my rankings, etc. This propagates all the way to UCF.

Using only games this year, UCF is rated 56.8 points better than FBS average, an unsustainable level of play even for Alabama.

Now, these team calculations get blended with other points based numbers from previous years to give you the public rankings on this site. A regression model determines the parameters.

Usually, the parameters work pretty well. Most of the remaining teams make sense. However, it completely fails for UCF. They benefit from playing only two games, and they get a big strength of schedule boost from the Maryland win.

Here’s the take home message: it’s not good enough to use just points based metrics to make football predictions.

Members of The Power Rank have access to ensemble predictions that combine these points based numbers with other calculations based on yards per play and closing market spreads. UCF has a much more reasonable rank of 47th in these numbers. The resulting prediction against Memphis also makes sense.

In addition, members have access to totals predictions for both college football and the NFL for this weekend, the first of the season.

To become a member of The Power Rank, click here.