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.
Last week, the model chose 2 games for upsets and missed on both of them, but fortunately, the Vegas lines slid before kickoff towards San Francisco and I picked up a game there by selecting Buffalo. So after losing 1 net game, my model and Vegas are now exactly tied with 128 correctly predicted wins (66.7%).
It was another good week against the spread, as the model went 10 for 15 and stretched the season win rate to 55%.
For week 14, the model predicts 3 upsets so far. However that could stretch to 4 as the CHI-HOU game is currently a pick ‘em.
I noticed that the drivers of 2 of these upsets (NE, WAS) is based on the performance of special teams. I’m not sure I want to trust a model that recognizes the best team by offense-defense match-ups, but then changes it’s mind because of superior special teams play. It’s something to look at in the off-season, I guess.
Out of the first 12 Colts games, the model has correctly picked the winner 9 times for a 75% accuracy rate. For the Raiders game, it gives the Colts a 62.5% win probability. I’ve done studies you know. 75% of the time the Colts win 62.5% of the time.
For the rest of the year, the model favors the Colts in 3 of the 4 games.
Given that current passing efficiency is exactly where I predicted it would be at the beginning of the year (0.16 EPA/db), the HIGH FORECAST numbers are pretty close to the rest of the year forecast with the exception of the Steelers.
At the beginning of the year, my model thought the Colts would be 4.6 point favorites over Pittsburgh, but now it thinks we are 0.9 point underdogs. We’ll see what that looks like in 3 weeks.