NFL DFS - 2020 Week 5

by Gus Elmashni

Happy Saturday everyone and what a great start I am having this weekend! Before I dive into my NFL DFS analysis, here are images of my MLB and NBA wins last night along with my rosters.

How am I building these lineups? There is some science and some art but of course some luck. I firmly believe that success in anything in life is a combination of being prepared and having serendipity on your side. For example, for the Rays - Yankees game last, I was fortunate to have Austin Meadows hit a homerun to tie the game but I was also well prepared. The predictive models on the Advanced Sports Analytics site (ASA) site strongly urged me to roster Meadows and they also encouraged a core of Gerrit Cole and Tyler Glasnow with one of them as Captain in DraftKings (DK) while the other was in the regular lineup. The link for ASA is below.

www.advancedsportsanalytics.com

That seemed risky since the game needed to be a pitcher's duel for that lineup to succeed and both pitchers ate up a lot of salary. Given that both the Rays and Yankees have strong bullpens and have strong managers, this risk was not a big one. To mitigate my risk, I entered two lineups: one with Cole as Captain and one with Glasnow as Captain. This strategy definitely paid off. The top lineup has a similar build but used some of the relievers (Aroldis Chapman and Peter Fairbanks). Again, you have to take some chances to win big in DFS and kudos to those who used such a strategy last night.

As for NBA, you don't need to be a rocket scientist to know that some combination of Jimmy Butler, LeBron James, and Anthony Davis is required, especially with two of them in the MVP and PRO spots on FanDuel (FD). The tricky part is determining which of those two you will roster since it's nearly impossible to roster all three and still have enough salary left for anyone else decent. That is where ASA's models help. By accurately predicting the minutes played and then determining the output, ASA was able to help me realize that Butler should stay in the MVP spot as he has been on fire since Game 2. It was a toss up between AD and LeBron but going with AD gave me enough salay cap space to identify 3 other starters with big minutes and decent usage: Tyler Herro, Duncan Robinson, and Kentavious Caldwell-Pope. For most of the NBA Finals last night, I was in the top 3 and at some points was in first place. Interstingly the 1st place lineup use Butler, AD, and LeBron while punting with Alex Caruso and JR Smith. Again, there is some luck involved but you have to love the art that comes into play when building lineups.

So how will we fuse art and science for tomorrow's NFL games? Now that we have in-season data for 4 games for most NFL teams, I have begun using my own models for NFL DFS. Attached are images of the outputs from the multiple linear regression model I built 2 years ago (for QBs and RBs). I don't have a WR/TE model yet but will be building one later this season with a focus on some advanced metrics such as Air Yards and aDOT.

FYI, here are the inputs I use for my models and I am happy to have a discussion with anyone on how each of those inputs are weighted.

  • DraftKings salary (logarithmic transformation for QBs and square root transformation for RBs)
  • Vegas point total (combined points for player's team and opponent)
  • Home/Away (positive weight for being at home)
  • First downs
  • 3rd down conversion rate
  • Completion percentage
  • % of total plays that are passes (for QBs)
  • % of total plays that are rushes (for RBs)
  • Turnovers (average of the giveways from the player's offense and takeaways from the player's opposing defense)
  • Sacks (average of those allowed by the player's offense and sacks earned from the player's opposing defense)

FYI, for the 4th, 5th, 6th, 7th, and 8th bullet points, I will average the results from the player's offense and the allowed values from the player's opposing defense. I then divide the players' outputs by their DK and FD salary to find the best values. The outputs are very robust for DK but can also be used for FD since the scoring is somewhat similar. For experienced NFL DFS players you need about 3 - 4x value in DK to hit the cash line and about 2.5x value in FD. The lineups I will share in this article are great for cash games but can also be used for GPPs. When playing GPPs, it is best to cast a wide net. Again, winning in DFS is part art, science, and luck.

The model idenitified 6 QBs worth starting and I believe that you start your core there. Pair your QB with pass catcher who has a high correlation. That is where the ASA comes into play. There is a Player Correlation app that I have found very fun to use when building my lineups. I will share correlations with DeShaun Watson (see below).

The correlations work both ways: correlating the QB with his teammamtes and the opposition. Most winning lineups will have a QB who pops off along with his favorite receiver and one player on the opposition. Again, we need to focus on strong scoring environments. After that, I like to find my defense. I know defense can be a crapshoot but the model I use has been reliable for the last 6 seasons. I use DVOA values from Football Outsiders to rate my offenses and defenses. I then compute the sum between the offensive DVOA and defensive DVOA for each game. Positive sums indicate a strong offense while negative sums indicate. A few years ago I built a linear regression model for fantasy defensive scoring which I still use today. I then divide the output by the DK salary. For this week, the Pittsburgh Steelers are the best value. There is the science bit. As for the art of it, I like Pittsburgh since they had a week off thanks to COVID-19 and the Philadelphia offense is injury riddled. If you can nail the right off defense at low ownership, you get the clear separation you need to make money in NFL DFS. Here are the results from that analysis.

I find tight end to be the biggest crapshoot for large slates. You could always pay up for an elite tight end like Travis Kelce but if he has a mediocre game, you are drawing dead. Sometimes you can correlate a strong tight end with your QB or an opposing tight end but the correlation must be strong with a decent sample size. Predicting which cheap tight end will do well is a tough task and is probably not worth the effort. This is the "luck" part of DFS that you will need on your side. Maybe on a short slate you can nail the tight end spot but you are better off focusing your efforts on RB and the remaining WRs. After all, they tend to get more touches than tight ends.

There are a few RBs I will consider for tomorrow. Ezekiel Elliott is always in the conversation given his elite status and high snap count. The problem is that the Cowboys defense is awful and have found themselves in shootouts every game so far this season. Maybe things will be different against the pathetic NY Giants but I am not so sure if I want to pay so much salary for a RB who can barely hit 2x value. With a few of the top RBs either injured or playing in one of the evening time slots, we will have to focus on the 6 - 7K range. Mike Davis, Clyde Edwards-Helaire (CEH), and James Conner all look like strong plays. Great game environments and they get tend to get 20+ touches per game. Raheem Mostert could be a low owned option given that he is coming back from injury and the 49ers play in the afternoon which sometimes gets overlooked compared to the earlier games. I also like James Robinson and correlate well with a Houston stack. ASA can also help with identifying strong RBs by using the Defense vs. Position app (see below).

The model loves Myles Gaskin due to his cheap price and high snap count but like Zeke, he is not quite hitting value. Joe Mixon went off last week but that was against Jacksonville. Could be a different story playing Baltimore and Mixon's history against Baltimore is not good. David Johnson can be a wise move in DK given his low price. You can do better on FD with more salary cap room but in DK, salaries are tighter and DJ correlates well with Watson and Will Fuller.

Finally, we will discuss our final WR spot. This is more science but some art. There is one WR who I feel is long overdue for positive regression: Marquise (Hollywood) Brown. Rain is in the forecast for Baltimore but it's light rain. Plus if the wind is light, sometimes a slick field is good for a WR, especially a fast WR like Hollywood. How are we suspecting positive regression? Air Yards and aDOT. Brown is among the leaders in those categories but has not experienced the strong DK and FD score outputs. Remember Will Fuller in Week 5 last year? He had a monster game with 200+ yards and 3 TDs. He was due and so is Hollywood for this year's Week 5.

Here are sample lineups for both DK and FD.

Curious to hear your thoughts. I will most likely run a few more lineups with other QBs and will share my lineups either late Saturday night or early Sundday morning both at Sportsmanac.com and Twitter. Feel free to chime in on my Twitter feed/post. Good luck with your builds and let's hope my DFS success continues on Sunday.