Earlier this week I introduced my wide receiver projections for 2016, which was developed using several machine learning algorithms.
Today I’m going to ensemble those projections with two more projections, those from our RotoViz Staff composite projections from the Projection Machine and the composite projections from fantasyfootballanalytics.net1 to find over and undervalued WRs — one per WR tier. We’ll start with the undervalued WRs.
Reminder — my machine learning projections only include players who saw 30+ targets last year or played in 8+ games, who have an ADP inside the top 90 WRs. That means no rookies and no players who didn’t play enough last year like Jordy Nelson. However, I still included them in the other projection systems for you to see, but won’t consider them for evaluation in this exercise.
Here are my machine learning projections (Machine), the RotoViz Staff composite (RotoViz), and fantasyfootballanalytics.net’s projections (FFA.net):
The ensembled projections are calculated in two ways. The first is just a straight average, and the second uses transformed z-scores to account for the fact that the projections themselves are in a slightly different range. For simplicity, we’ll use the average projections the rest of the article, but the EnsembleZ projections should be slightly better in the long run.
Undervalued WR Candidates by Tier
WR1: Allen Robinson — Positional ADP: 7 — Composite Rank: 5
Allen Robinson looks to be the most undervalued from the upper echelon of WRs, coming in fifth in both ensemble methods but only being drafted as the WR7. Robinson may regress a small amount, but his yards per target over expectation (YPTOE) was actually negative, meaning there’s room to imrpove in that area. He also had the second most air yards, which is a main reason the machine learning model likes him. Air yards are a sticky stat from year-to-year, so if that air yards volume remains and his catch rate improves, we’re talking about a player who could repeat as a top six season-long WR.
WR2: Julian Edelman — Positional ADP: 23 — Composite Rank: 20
Edelman hasn’t been getting much love this offseason for two reasons. First, he’ll have four games without Tom Brady, and second, he’s coming off a foot injury. However, all indications are he will be good to go for Week 1. In this case, Edelman is boosted by the machine learning model, as he fits the mold of a high producing receiver with a very short air yards per target (along with Jarvis Landry, Randall Cobb, and Golden Tate, to name a few). The machine likes these receivers’ consistent volume, which is king. While most of the WR2s are fairly valued, if Edelman can mostly reproduce with Jimmy Garoppolo what he did with Brady the last two years, he’s the undervalued man in the bunch.
WR3: Jordan Matthews — Positional ADP: 29 — Composite Rank: 24
Matthews is a guy I’ve come around to after a great Slack chat debate between myself and a few other RotoViz colleagues on if we’d rather have Matthews or Larry Fitzgerald. According to the ensembles, the Matthews crew won — barely! Matthews is poised for a breakout season, and my young WR model also likes Matthews to shine because of his insanely high collegiate market share, which still matters for third year receivers. Oh, he also has some interesting numbers when compared to another guy on this list…Allen Robinson.
WR4: Stefon Diggs — Positional ADP: 41 — Composite Rank: 32
The machine learning model is agnostic to QB play, so the projection for that stands independent of the serious knee injury to Teddy Bridgewater. My best guess is Diggs’ ADP will fall a slight bit, but it will likely be mirrored by slight drops in the RotoViz composite or the fantasyfootballanalytics.net composite. With both ADP and ensembled projections taking a slight hit, Diggs would still be a crazy value. I don’t need to rattle off the myraid of reasons we love Diggs at RotoViz. I’ll just link you to the articles here, here, here, and here.
Oh, and I don’t believe Shaun Hill, or some of the names floated around as trade targets, would hurt Diggs too much. Here’s an AYA graph comparing Bridgewater to Hill and the others.
Ignore Hill’s 2015 — it came on only seven pass attempts. But his 2014 season with the Rams was when he started eight games and lands smack in the middle of this bunch.
WR5: Steve Smith — Positional ADP: 60 — Composite Rank: 47
We’ve already identified Smith as a value, and he’s going to be the WR1 on this team when all set to return in Week 1. Last year, Smith was on a blistering pace which would have put him 13th in total air yards over a full season. Yet, he hovered right around his career average in efficiency, so he wasn’t boosted unnecessarily by efficiency. If he can retain 60 percent of that volume, and his career average in efficiency, he’ll be a WR4 for a WR5 price.
WR6: Pierre Garcon — Positional ADP: 64 — Composite Rank: 52
While Garcon might not have the most upside in the WR6 tier, he holds the most value according to the ensemble rankings. Garcon saw the 35th most air yards last year and was near league average in both YPTOE and TD rate. The addition of Josh Doctson is concerning, but Docston isn’t expected to be ready for Week 1, and could miss more time than that. Brian Malone expects Garcon to maintain his 20 percent market share, and the models agree. He’s not young, or flashy, but he represents a ton of value.
WR7: Jermaine Kearse — Positional ADP: 75 — Composite Rank: 66
I’m snapping up all the Kearse I can at his current price. Let’s compare his 2015 to that of teammate Tyler Lockett.
Remarkably similar. He trailed in catch rate and TD rate, but led in YPT. Combining the two, along with game situation, he actually was nearly equal to Lockett in YPTOE. Add in a likely unsustainable TD rate for Lockett, which was the 96th best out of 1350 WRs with 50+ targets in a single season since 2000, and we have a situation where the two are much further apart in ADP than they probably will perform on the field in 2016. Kearse is still the WR2 on the Seahawks depth chart, and has played and practiced as the WR2 in the preseason.
To be clear, I’m not arguing for Kearse over Lockett; I’m just saying Kearse should be much closer to Lockett in ADP.
You can use the Diff and RelDiff columns to find more undervalued candidates. Some good ones include Emmanuel Sanders, Michael Crabtree, Torrey Smith, Vincent Jackson, Mike Wallace, Anquan Boldin, and Robert Woods to name a few.