Kevin Cole's 2018 NFL Betting Model: Week 4

Along with working extensively on NFL projections this season, I’m also putting the data to use projecting game outcomes and scores. We can, of course, apply these projections against the going betting spreads and over/under totals to see where there might be value.

In addition, this information can provide value in DFS by finding teams who are likely to outperform their point totals implied by the spreads and over/unders.

Each week this season I’m going to post the model’s spread and total picks on with the lines as of Thursday and Sunday.

Methodology

The betting model incorporates roughly seven years of historical data to train the model and it was tested on the 2016-2017 seasons, yielding hit rates of roughly 60%. While the model has shown promising results, it is still a work in progress and needs additional testing to boost out-of-sample performance. The numbers below are best viewed as a tool, not the definitive answers.

The model incorporates multiple rolling periods or opponent-adjusted data for offense and defense, adjustments for pace, and adjustment for quarterbacks. Non-quarterback additions and injuries are not built into the model. If you believe Khalil Mack is worth multiple points per game for the Chicago Bears defense, make adjustments accordingly.

The model also looks at QB and team stats together, which can hold down the numbers for teams like the Chiefs who have strong QB play, but not much history.

The model does not yet account for weather, which isn’t a big concern this early in the season.

“Spread Pick” Those highlighted in green means the favorite is projected to beat the spread, red if the underdog is projected to beat the spread. “O/U Pick” is green for games predicted to go over, red for games predicted to go under.

Week 2 Review

Last week, the model had a rough week at 3-9 based on Thursday lines and 3-8 on Sunday, driven by difficulties projected high-total games. Games with a total over 50 points went over 4-to-1 last week.

It’s possible there has been a fundamental shift in scoring that will mean under plays on high total games won’t be successful this year, but I’m going to stick to using the model results and not make reactive adjustments based on a few weeks of play.

Here are the number so far this year

betting_model_week4_results

Week 4 PicksPicks as of Sunday at roughly 10 a.m. ET

betting_model_week4_sunday_plays

Changes from the Thursday lines include Jaguars falling off, with the Cowboys and IND-HOU Over becoming plays. It was actually an oversight that I left the Cowboys off of the chart on Thursday, but you can’t make changes to the chart after the fact.

The purpose of posting these picks is to give you an idea of what a cold-blooded model thinks about the games, but we also have to keep in mind that it has blind spots. One of those is injuries to non-QBs, so I’m not particularly confident that the Falcons will cover after being decimated on defense. The spread has been moving against them all week, and the total has also been on the rise.

The model continues to view the Packers skeptically, and picks against them for the second straight week. This is also two weeks in a row of picking the Bills who have performed much better with Josh Allen under center.

The Saints have an unfavorable spread to beat at -3.5, but the model still thinks there’s value and sees them as one of the best teams in the NFL.

Here are the implied totals from the model versus betting lines.

betting_model_week4_implied_totals

Picks as of Thursday at roughly 12 p.m. ET

week4_betting_model_new_thurs_picks

No play for tonight’s matchup, but we have lots on the schedule for the rest of the week. Again, the model is recommending going under on the Chiefs, which could be right or a blind spot in the model seeing Pat Mahomes as a limited experience QB that is likely to regress, instead of the uber-talent he may be.

I’ll include more commentary for the rest of the games on Sunday morning.

About the Author

colekev
Kevin Cole (colekev)

Kevin Cole previously worked on Wall Street as an equity and credit analyst before transitioning to data-related sports analysis. For a number of years, Kevin has specialized in creating predictive sports models that includes published work at Pro Football Focus, Rotoworld, numberFire, and RotoViz. He is now the Director of Data and Analytics at RotoGrinders.