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.
Like the Colts, my model faded in week 16. After the Vegas lines settled, the only upset pick I had was TEN over GB by 1 and I was only off by 27 points. So, dropping another game to Vegas puts me 3 games back and out of reach for the season title. Damn you Vegas. Damn you to hell.
Against the spread, the model was 6 - 10, by far the worst week of the year. If it were an NFL team, that record wouldn’t even be good enough to win the NFC East.
I have to say, though, I am quite happy with the season results (so far). A 66.4% win prediction rate is right on track with what I expected and I think with a bit of work, I could raise that a few points. Against the spread, a 54.9% win rate is much higher than I would have guessed. There’s a lot of luck involved, but there seems to be a decent signal in there.
Absolute Point Accuracy of 10.05 shows that my average spread is about 0.2 points farther away from actual scores than Vegas spreads (9.86), which isn’t bad at all. The Net Point Accuracy of 0.12 shows that I have almost no bias in my spreads against home/away teams, but Vegas underestimates the visiting team performance by about 1.2 points per game, so there is some leverage there.
This week should really challenge the model, as some teams will be resting starters and I have no mechanism to account for that (other than for back-up QBs that have established historical data).
However, even with those variables, my predicted spreads are pretty close to Vegas. The only upset I have is that even with Patrick Mahomes on the sideline, I have the Chiefs squeaking by the Chargers.
There are still a lot of teams that have not announced their starting QB — I’m looking at you Buffalo — so these numbers are subject to change prior to kick-off.
I have the Colts at a 77% chance of beating the Jags, which brings the final expected season win total to 10 (9.7). So if the Colts lose in week 17, my model will have nailed total wins for the year. Here’s to hoping that I am off by 1.
If the Colts beat the Jags, then my model will have correctly predicted 12 of 16 Colts game outcomes, which is pretty good. If I could average that, I would start shifting my retirement money to football games.