Week 5 NFL model projections
This is part of a recurring series on VSiN.com in which VSiN host Jonathan Von Tobel tracks the progress of a model he created for the NFL season
Week 4 Best Bets: 1-4 | Overall Best Bets 12-8
Week 4 Overall: 6-10 | Overall Record: 37-26-1
Week 4 Recap
Last week was an interesting one for Jon Von Model. As mentioned in the last column, the projections were extremely tight with the market, as seven projections were within a point of the betting market. That accuracy did help with the results though, as our contest card went 1-4 and overall edges were 6-10 for the week. Quarterback changes also hurt the overall record of edges, as the initial projections did not account for the absences of Deshaun Watson or Jimmy Garoppolo.
Four weeks into this experiment and the model is still outperforming expectations, even coming off of an abysmal week. The time is coming for the model to run completely on data from this season, but for at least one more week we will run on a version of the model that got us to this point.
Week 5 Adjustments
This week was the most challenging week when it came to adjustments. Overall edges have hit at a 58.7% clip and the projections are in line with the market, but a bad week seemingly shook my confidence in JVM. When I entered the games for this week my confidence was shaken a bit more.
In the initial projections two things immediately stuck out: Jon Von Model was still too high on Atlanta despite the addition of Desmond Ridder a couple weeks back, and New England was coming out too strongly rated.
JVM’s initial projections had Atlanta favored by 8.68 points, and New England was favored by 8.23 points. Perhaps those edges will prove to be true on Sunday, but I thought it warranted an adjustment.
Remember, this version of the model is still using statistics and data from last season, so I decided to overhaul both teams. Not only did I make sure Ridder’s statistics were up to date, but I also made sure to include updated team statistics for the Falcons as well as the Texans. The adjustments came back with a projection much more in line with the market this week, as you will see shortly.
I followed the same process for the Patriots as well. One of the biggest changes that needed to be made was to Mac Jones. As someone who believed that Jones would play closer to the level of his rookie season I had used his statistical profile from that year when I built the model in the summer. Those numbers were still being attributed to Jones, when the reality is that he has not been very good. He is among the lowest graded starters in PFF passing grade - one of the metrics JVM uses - and he has the second-worst turnover worthy play rate as well. Those changes brought that projection back in line with the betting market as well.
Green Bay also needed an adjustment. They were favored to win by 6.32 points when I ran the initial projections, but that too warranted an adjustment I believe. JVM has been high on the Packers since the start of the season, and I believe the root cause is some of the team statistics are still inherently tied to Aaron Rodgers. Green Bay also received a full makeover this week, and the result was a number that favored Las Vegas.
Each week I also update another version of Jon Von Model which is built entirely on 2023 statistics. The numbers for each contest are getting more in line with every passing week, but the outliers prove that the model needs more data to work with. I expect that by Week 7 the projections will be turned over fully to that version of JVM.
Week 5 Projections
The elephant in the room is the first projection on the board. Jon Von Model has Washington projected to win by 14.81 points on Thursday night. That is a massive 9.31-point edge for the Commanders, depending on where you shop.
After submitting the projections to our graphics department I made sure that Justin Fields’ numbers were up to date, and that did make a difference but it only lowered the projection to 11.37 points. The way JVM is built - based heavily on offensive statistics and few defensive metrics - a team like Chicago is destined to be at the bottom of the projections. In fact, four of the five weeks this season the Bears’ opponent has been one of the five biggest edges. The only exception was last week against Denver, and still the model had the edge toward the Broncos.
The same problem continues to happen at the top with Kansas City. Once again the Chiefs appear on the card as one of the biggest edges of the week. Kansas City has covered two of the three times it has been included on the card, but this week will be spent figuring out why the Chiefs constantly come out on top. It is one thing to have Kansas City as the best team in the NFL, but the edges should not be so lopsided in their favor.
There was a similar issue with Cincinnati - it was a contest play last week if you don’t remember - but an adjustment this week consisted of updating Joe Burrow’s statistics as well as some team adjustments. Those same adjustments are not moving the Chiefs down at all, so there needs to be a deep dive done this week to figure out the issue.
Moving on to the important stuff; the contest entry. Here are the five biggest edges of the week, according to JVM:
Mitch Moss and I spoke about using the Commanders as a contest play despite us wishing to pass on Thursday games altogether, so for now we will include them as one of the five plays for this week.
One thing to note - much like last week - is that the margins are getting tighter. Jon Von Model has edges of at least 4.28 points on the top three plays, but the other two are only 2.32 points and 2.28 points respectively. Should we decide against using the Thursday night game that would mean three of our contest plays have edges of less than 2.5 points.
Whenever I get to interview established bettors in the space who have systems like this I always ask what minimum edge do they need to consider a game playable, and the answers vary. In a market like the NFL I would want a larger edge than just 2.32 points, but we need five plays for the contest so here we are.
This will be a big week for Jon Von Model. Not only because of the week it had previously, but because we are getting closer to moving off this current version and on to the 2023 version of the model. It will be a good test on how deep into a season can you rely on last year’s data before moving on from it.