2017 Air Yards Projection Countdown | Mike Wallace – WR 33
Now that the draft is over I’m excited to report that I’ve finished my air yards model for NFL receivers.
Well-validated models are the best and most unbiased forecast of NFL receiver production there is.1 However there are things that they cannot capture that are important to consider.
Changes to team composition, injuries, coaching changes, and more can affect the accuracy of a pre-season model that is based purely on numbers. Finally, many models, including mine, do not give more than an expected value forecast for a player. Point estimates are great, but it’s also crucial to understand the range of outcomes.
In this series, I’m counting down the model’s top 36 wide receivers and publishing forecasts along with a discussion of the other important factors the model doesn’t know about. Full projections will be released next month.
Air Yards Model Projection
Projected PPR Points per Game: 14.7
Projected WR Rank: 33
Actual ADP2: 52
Model Difference: + 19
Team Personnel Changes
Earlier this offseason Charlie Kleinheksel and Kevin Cole used different methodologies to arrive at the same conclusion: Baltimore lost a metric shit-ton of receiving talent this offseason. Steve Smith retired, Justin Forsett retired, Kyle Juszczyk left for San Francisco, and Kamar Aiken signed with the Colts.
The fascinating thing is that the Ravens then proceeded to take absolutely zero receivers in the 2017 rookie draft. The only offensive players they ended up drafting were Guards in the fourth and fifth round.
This leaves a gaping hole of opportunity for presumptive starting outside receivers Wallace, Breshad Perriman and slot-man Michael Campanaro.
With no real competition for targets added to the team, it seems an easy projection to bump Wallace up to 21 percent target share.
The ARIMA projection3 for the Ravens shows a team that loves to throw the ball early and often. But will it continue?
Floor: 620 attempts
Ceiling: 720 attempts
Expected: 670 attempts
Bovada Sports Book has the win total for the Ravens at 9. Their seasonal game script was fairly neutral in 2016 so based on these two pieces of information we should probably expect their pass/run split to remain similar to 2016. That might not turn out to be the case though in 2017.
Despite a negative trend in plays per game over each of the past three years, the Ravens have continued to rely more and more on the pass. This trend tracks almost perfectly with seasonal game script. As the team got worse, they passed more but ran fewer total plays.
The obvious explanation is that that teams that are behind pass to come back, but there may be other secular factors at play (including the overall league trend toward more pass-happy offenses).
The net of all this is that if the Ravens do improve and become an above-average team, we should probably expect fewer pass attempts and more total plays run.
I’ll go with 650 attempts expected, taking the under from the ARIMA model on the assumption that the added volume of plays won’t cancel out the drop in passing attempts completely. This leaves us with the following range of outcomes for Wallace.
Floor: 130 targets
Ceiling: 151 targets
Expected: 136 targets
Wallace has been a decently efficient receiver throughout his career. What is slightly concerning is that Wallace’s career catch rate is below average on the high-value deep targets that quarterback Joe Flacco has a reputation for throwing well.
When you combine his catch rate with Yards After the Catch you get a metric I call RACR. Using it, you can see that in general Wallace is a league average receiver at almost every depth of target. He is quite good on shorter passes though, which should help him entering into his age 30 season.
His career aDOT is 13.8 and his career RACR is 0.62. Let’s calculate his receiving yards.
Floor: 130 targets * 13.8 aDOT => 1794 air yards * 0.62 RACR => 1112 receiving yards
Ceiling: 151 targets * 13.8 aDOT => 2083 air yards * 0.62 RACR => 1291 receiving yards
Expected: 136 targets * 13.8 aDOT => 1876 air yards * 0.62 RACR => 1163 receiving yards
Wallace’s career catch rate is 57 percent. His career catch rate is impacted by the fact that he’s seen lots of deep targets. If you adjust for depth, he’s caught just about as many balls as you would expect an average NFL receiver to catch. Let’s calculate receptions.
Floor: 130 targets * 0.57 Catch Rate => 74 receptions
Ceiling: 151 targets * 0.57 Catch Rate => 86 receptions
Expected: 136 targets * 0.57 Catch Rate => 78 receptions
Finally let’s calculate TDs. Wallace’s career TD Rate is 6.2 percent.
Floor: 130 targets * 0.062 TD Rate => 8 TDs
Ceiling: 151 targets * 0.062 TD Rate => 9.3 TDs
Expected: 136 targets * 0.062 TD Rate => 8.4 TDs
Wallace’s final back of the napkin projection is very close to the air yards model, and there is substantial upside.
Let’s compare it to the Sim App:
In a departure from the previous trend in this series, the back of the napkin calculation is substantially more bullish on Wallace than both the Sim App and the air yards model. The range of outcomes in the sim app are also quite tight, which is fascinating. It implies that Wallace is a fairly low risk pick at this point in his career.
With an ADP of 52 and a model ranking of 33, Mike Wallace is the first large departure the model has taken from the wisdom of the crowd. Our role as analysts is to try and determine if we should give more weight to the machine or the men. In the case of Mike Wallace I’m quite comfortable in saying that the market simply has him mispriced.
The things the model doesn’t know about Wallace all point towards an even higher ceiling than it gives him credit for. Wallace is a huge bargain at his current price, and I expect the market to correct on him very soon.
Subscribe for a constant stream of league-beating articles available only with a Premium Pass.
- The air yards model combined four separate machine learning models into one ensemble model. The models used included a Bayesian neural net, a nearest neighbors algorithm, a random forest algorithm and a tuned generalized linear model. Preprocessing steps included data normalization (centered and scaled), the elimination of near zero variance predictors, and principal component analysis transformation. The model data used were 838 observations of 35 predictors, including age, weight, previous season player volume and efficiency data, 3 year weighted player efficiency data, and previous season team passing volume and efficiency data. The dependent variable was fantasy points per game. The data were split 80/20 into training and test sets. 10 fold cross validation was performed. The measured R^2 on the held out data was 0.66. (back)
- as of early May 2017 (back)
- these are posted for every NFL team on airyards.com (back)