clock menu more-arrow no yes

Filed under:

2020 NFL Win Forecasts: Week 16

NFL: Pittsburgh Steelers at Cincinnati Bengals Joseph Maiorana-USA TODAY Sports

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.

All data used originates from Pro Football Reference and nflFastR.


RESULTS

There was a lot of movement in the data after I published my picks last week, so I want to address the changes.

  • I found an error in my week 15 special teams data that ended up shifting my pick from MIN to CHI
  • I had not accounted for Drew Brees returning and when I added his data back, my model picked NO over KC.
  • The line shifted away from Miami to New England, which made that an upset pick for the model
  • An error in my week 6 data took away a straight up win from me when corrected

The result of all that is that I was 3 games down entering the week and picked 2 of 3 upsets and am now sitting 2 games back of Vegas with 2 weeks to go. Yikes.

On the other hand, the model continued it’s outperformance against the spread picking 10 winners correctly for the 4th week in a row. That ups the season total to 56.4% against the spread.

Net point accuracy against the spread of 0.17 per game is just insane.


PREDICTIONS

It will be interesting to see how the model fares in the last few weeks with teams resting starters or having nothing to play for. One would think, historical numbers would not predict well in those situations. However for now, the model is picking 2 upsets: Carolina over Washington and Tennessee over Green Bay (booooooo!).

There’s a pretty big discrepancy in the Kansas city spreads. That is likely because Mahomes’ last 3 weeks have not been great EPA-wise and so my model is probably overweighting recent performance (or is my model right and Vegas is under-weighting?).


COLTS SEASON

The model predicts 13 binary wins in the regular season, which I am willing to bet the Colts won’t reach. So far this year, the model has predicted 11 of 14 game outcomes for the Colts, missing only on JAX, CLE and the first TEN game.

That’s actually a bit better than I thought it would do, but there’s still 2 games left. The model likes the Colts in those games, but only barely so against Pittsburgh.