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.
Week 8 results rivaled a well executed coin flip. The model picked 7 of 14 winners straight up and 7 of 14 against the spread. It also went 1 for 2 in predicted upsets and so neither gained nor lost ground against Vegas, maintaining a 1 game edge on the season.
My spreads were 2 points low on average, which brings my season average deviation from actual spreads to 0.0. You can’t get much better than that and it’s a big driver as to why the model is almost 55% against the spread on the year.
Also, the 67.2% accuracy in straight up wins is 1 game higher than the expected accuracy. So far, this model is performing way, way, waaaaay better than I thought it would. I’m sure a big part of that is randomness, but it’s still fun to watch.
Week 9 is the first game where my model does not predict a Colts victory. It gives Baltimore a 58.9% win probability, which translates to a 2.8 point spread. That is in line with Vegas, but my gut tells me it won’t be that close.
The data predict 2 upsets next week: the Saints over the Bucs and the Giants over the football team that dare not speak its name. If I lose both of those, Vegas will re-take the lead. Can you feel the tension?
Model spreads compared to Vegas spreads are similar except for a couple of games. Obviously the upsets have big variance, but even though I agree with Vegas on the winner of DAL-PIT, I have DAL as 7 point underdogs and that probably should be more. My model doesn’t know what is going on at QB in DAL. Same with SF-GB.
So, I still need to develop a replacement QB logic. If I update anything before kick-off, I will post a comment to document it.
The win against Detroit boosted the expected wins total to 10, the highest mark it has been all year.
Before the season started, I predicted that Philip Rivers weighted average EPA/db would at least match his 2017-19 averages and this week he has achieved that. That’s why the forecasted wins match the ‘HIGH FORECAST” predictions set at the beginning of the season.
We’ll see if he can maintain that level of passing efficiency and if that actually translates to 10 wins.