Greetings from the secret, WiFi-enabled Razzball bunker! It has taken a while but we finally got our robots to broaden their heretofore MLB-limited expertise and start cranking out NFL in-season projections for both season-long (Standard, 1/2 PPR, PPR) and DFS.

It has been a mad sprint to being ready opening day so I don’t have all the documentation to share but hoping the below summary will suffice for now.

Before I lose those of you who have no interest in how the sausage is made, we’re making both the season-long and DFS projections free for the first 4 weeks of the NFL Season.¬†Below¬†are the links (also can be found under Tools in the top menu). Player pages should be up shortly (ETA end of next week).

Pigskinonator – In-Season Projections

DFSBot – DFS Projections

Still here? Here’s a quick bulleted overview of how we created the Pigskinonator:

  • Start with a massive file of play-by-play NFL data (over 10 years’ worth)
  • Use Vegas over/under and spreads as the foundation for the team projections.
  • Use past game data for players to estimate both their opportunities (e.g., snaps, rushes, targets) and skills as measured by ratios (yards per rush/reception, completions per target). Test biasing factors (e.g., domes, weather, home/road) and adjust the data to be as ‘neutral’ as possible. Regress player data to league-average for the position to deal with the inherent sample size issues.
  • Adjust the player data to fit the team projection – e.g., if I’m projecting 42 pass plays for Pittsburgh, I need to make sure the sum of all receiver targets is 42.

There is much more to it but it’s 4AM. Let’s switch to a Q&A format – if you have questions, ping me in the comments of either this post or one of the projection pages.

If I have learned anything from building and maintaining our MLB projections, it is that the first release is just a milestone. The projection methodology and the usefulness of the tools will be better a month from now and it’ll be even better for 2017. How much better depends on all of your feedback.

So, please, dig into the projections. Find the anomalies. Tell me what stats/reference data to add. Help me help you!

 

  1. MattTruss

    MattTruss says:
    (link)

    I’ve been waiting for you to take your projection talents to NFL Rudy! Can’t wait to check this out, optimizing Vegas over/unders is something I’ve been using to create my own projections in past seasons, this is going to save me so much work! Things look great at first glance, Team Defense pricing is whacky for the DraftKings projections…minor. Appreciate all you do Rudy, excited to see how the season goes.

    • Thanks man. I think I got the DK pricing thing resolved but not 100% sure.

  2. For an Armenianless Vacation Come to Akron says:
    (link)

    sweet jesus nice.

  3. Aubrey Plaza's Pillow says:
    (link)

    now let’s see how it did vs spreads and then in actual daily results against said spreads (i’ll use closing spreads)

    1. SEA -10.5 (machine says -10.1), machine goes 0-2 (against real spread and it’s own)
    2. ATL -2.5 (machine says -2.9) machine goes 0-2
    3. MIN -2.5 (machine says -2.2) machine goes 2-0
    4. PHI -4 (machine says -3.9, and interesting as this game opened as PHI by -7, then moved by market down to machine’s level, but that could’ve been caused by bradford trade) machine goes 2-0
    5. CIN -1 (machine says -2.4) machine goes 0-1-1
    6. NO -3 (machine says -1.2) machine goes 0-2
    7. KC -6.5 (machine says -6.6) machine goes 0-2 (barely)
    8. BAL -3 (machine says -3.2) machine goes 2-0
    9. HOU -6 (machine says -5.9) machine goes 2-0
    10. GB -3.5 (machine -4.7) machine goes 2-0
    11. DAL -1 (machine says -.1, closest thing it’ll have to a flip yet not a flip) machine goes 0-2
    12. IND -3 (machine says -3.7) machine goes 0-2
    13. ARI -9 (machine says -6) machine goes 0-2
    now a record of 5-7-1 against closing spreads and a record of 5-8 against it’s own projections isn’t great, BUT if we ONLY bet in instances of it having a 1.2 point edge or greater when comparing it’s projections vs the spreads we’d have bet ONLY these:
    NE +9
    GB -3.5
    OAK + 3
    CIN -1
    and we’d have gone 3-0-1, which is awesome (small sample of course). i myself would have gone
    6-6 (i’m throwing out the JAX/GB game as earlier i would’ve gotten a JAX +4.5 not +3.5, and i’m not gonna claim one way or the other which way i’d have gone, since i don’t know and in this instance think it’s too close to claim i’d have picked either way). i’m getting very interested in this thing.

      • Aubrey Plaza's Pillow says:
        (link)

        @Rudy Gamble: i was going off one book shop so that explains why they were usually nearly identical. and the particular market of that shop probably explains any differences. if true, this explains why these 4
        NE +9
        GB -3.5
        OAK + 3
        CIN -1

        were off at least 1.2 points from your projections. it’s just that the market for those games was farthest from vegas in their closing numbers.

        • Aubrey Plaza's Pillow says:
          (link)

          @Aubrey Plaza’s Pillow: so you probably did update them if you think you did.

        • @Aubrey Plaza’s Pillow: no idea. getting the odds from a feed provider (Fantasy Data). Let’s watch during Week 2 and see if it’s tracking correctly early in the week.

          • Aubrey Plaza's Pillow says:
            (link)

            @Rudy Gamble: right now it’s showing some weirdness on the LA-SEA game, pretty sure no matter how bad LA is they can’t get -4 points.

Comments are closed.