Identifying Fantasy Football QB Regression Candidates

By Kurt Bullard

When it comes to approaching fantasy football drafts, the ultimate goal is to try to find value—having to pay less than what one should for an asset. One popular way to do that is to try to identify players who are bound to “regress to the mean” after stellar or subpar campaigns. One can do this relatively qualitatively; for example, Blake Bortles is a prime regression candidate after throwing the second-most touchdowns last year, and a lot of the speculation is because people don’t believe that he’s talented enough or surrounded by good enough players to sustain such an output. The obvious downfall to such analysis is that it still relies on the eye-test. Regardless of personal beliefs, Blake Bortles could actually be in the right position and possess the right tools to repeat his 2015 output.

I wanted to see if there was a way to identify regression candidates using a more-strictly quantitative method. For the purposes of this article, I am only looking at the quarterback position to begin with. To move forward with the analysis, I looked at all pass attempts from the 2015 season for all QBs and looked to create a model to predict the point value of each pass based on the following factors:

  1. Down

  2. Field Position

  3. Yards to First Down

I used a general additive model and settled on regressing the fantasy points created by each throw against the interactions of down and distance as well as the field position. (It does not include points based on runs, so the likes of Cam Newton are disadvantaged in this analysis.) From there, one can find the expected points created by each quarterback throw and its residual. It’s possible that quarterbacks that severely strayed from their expectation are regression candidates. The following is a table of the sum of the expected points for each quarterback based on the context of their throw, as well as by what extent they under- or over-performed their expectation:

The biggest candidate for backwards regression is Russell Wilson. According to the model, @dangerusswilson overperformed by 86 points last year, which was 30% of his season total. Derek Carr and Tyrod Taylor are also on the radar after surprisingly solid campaigns. Interestingly, Wikipedia aficionado Bortles did not over-perform his expectation by a large extent, only beating it by 29 points.

On the other side of the spectrum, besides some iffy quarterbacks who are no longer starting for NFL teams, it looks like Joe Flacco, Aaron Rodgers, and Matt Ryan may present value for fantasy owners, as they underperformed their expectations by double-digit percentages. Rodgers was playing with an injury-depleted receiving corps, while Matt Ryan only put up 21 TDs in addition to his 4,500 yards. Similarly, Flacco threw less than a touchdown-and-a-half per game in his 10-game stint last season.

This analysis does not necessarily guarantee that Russ or Carr will regress—it could be that they’re such great talents that they can sustain such performances. But it does suggest that they may be candidates for less impressive fantasy seasons.

About the author

harvardsports

View all posts

1 Comment

  • This analysis does not necessarily guarantee that Russ or Carr will regress—it could be that they’re such great talents that they can sustain such performances. But it does suggest that they may be candidates for less impressive fantasy seasons.

    What are the chances you could get a couple more seasons of data? That might help to measure the repeatability of over/under performance.

Leave a Reply

Your email address will not be published. Required fields are marked *