This is an exercise in building a simple blind predictive model based on a team’s historical stats in 4 areas: pass and rush efficiency, and avg EPA for special teams and penalties. The season long experiment will track how well future games are determined by past statistical performance and compare that record against Vegas betting lines.
In week 5, my model went 10 of 14. It predicted 2 upsets and hit on both of them (CHI, TEN) gaining 2 games on the betting lines. However, right before the Colts game, the line shifted to the Browns and I had to give one of those games back.
On the year, I am 51 of 77 (66.2%), which is 2 games behind Vegas and about 1⁄2 a game up on expected accuracy.
For week 6, I only disagree with Vegas on 1 game. The model thinks the Bears will win one on the road against Carolina.
(EDIT: table updated to reflect NFL schedule changes)
The overall spread differentials are pretty close, except my numbers have the Titans crushing Houston. I only use point spreads as a gauge, since the model doesn’t specifically predict points, rather just win probability.
I am using a separate formula outside of the model to translate that to spreads as people are far more familiar with that measure, but my spreads should be in the ballpark. If they aren’t then it’s a flag for me to check under the hood . . . like TEN-HOU.
Last Sunday’s performance pushed the Colts’ expected win total down to below 9.5, but it still only has 3 upcoming games with a <50% chance of winning. I’m sure that will change as the year progresses.