DFS Turnaround - Week 18


This content is not available due to your privacy preferences.
Update your settings here to see it.


The dynamic game of Daily Fantasy Sports (DFS) requires much more than simply knowing the sport for which we're entering contests to be successful. We must be adaptable, precise, and open to learning from previous endeavors, the latter of which will be the primary focus of this weekly written piece. Game Theoretic methodologies will allow us to analyze and dissect the previous week's winner of the largest and most prestigious Guaranteed Prize Pool (GPP) tournament on DraftKings – the Millionaire Maker. These same tenets of Game Theory, which can most simply be explained as the development of decision-making processes given our own skill and knowledge, assumptions of the field based on the cumulative skill and knowledge of others playing the same game, and the rules and structure of the game itself, will allow us to further train our minds to see beyond the antiquated techniques of roster building being employed by a large portion of the field. Approaching improvement through these methods will give us insight into the anatomy of successful rosters and will help us develop repeatably profitable habit patterns for the coming weeks.

We're going to change things up a bit this week as the regular season comes to an end (and with so much variance attached to the Week 18 slate) and examine some of the things we've learn, reflect on the process of building a positive expected value (+EV) DFS roster, and apply those lessons to the short slates that the major DFS sites have provided to us for the Super Wildcard Weekend.

It’s NFL season and that means you’ve got a chance at $100,000 with Sunday Night 7. Predict what will happen on Sunday Night Football and watch along to see if you’ll be the next big winner. Click here to get started with the NBC Sports Predictor app.


Lessons Learned This Season, Applied to Super Wildcard Weekend

Optimal vs. Winning Lineup

Put most succinctly, the difference in scoring between the optimal roster and the roster that took down the Millionaire Maker on DraftKings appears logarithmic based on the size of the slate. As in, the smaller the slate (in number of games), the closer to the optimal roster you would need in order to ship a large field GPP; the larger the slate, the further from optimal you can be and still win. This should make sense fundamentally as there are simply more variables introduced with more games, but the distinction bears repeating as many DFS players don't alter their decision-making processes based on this basic information. This could not be more pertinent to the contests available on the Super Wildcard Weekend slate - DraftKings has broken the slate up into three pieces, with the largest contests available for the two games on Saturday and the three games on Sunday, but also offering a "full slate" look consisting of all six games (Monday included). As such, the process of constructing rosters should theoretically be different for each respective slate, with more reason to embrace variance the fewer games are present on a given slate.

Relative Decision-Making

The interesting theoretical component at play for Super Wildcard Weekend is the ability to gain more information with each game played due to the fact that each individual game will end prior to any other game kicking off. While the field understands the basic tenets of the late swap functionality on DraftKings, few actually utilize it nor realize that not all decision for the slate need to be made prior to the first game kicking off. What I mean by that is this - the process of Game Theory fundamentally describes the action of collecting information (projections, optimizers, building rosters, etc), making assumptions regarding how you expect the field to behave (ownership projections, analyzing how the chalk comes together, analyzing the state of each slate, etc), and formulating a plan to maximize expected value considering the previous inputs. It is a process of tailoring one's decision-making process to maximize expected value over time - that's it.

With that understanding, would not the decision-making process change as more information is added to the equation, and would we not be able to collect more information after each subsequent game? The answer is a resounding "yes," yet only a small portion of the field is thinking through these second and third layers to DFS play! Relating that thought process to the previous learning point, we can then bias our future decisions towards more or less variance based on the outcomes of the preceding game(s), relative to the size of the slate. Here are a few questions to ask yourself at each new decision node (before each game on the slate this weekend) in the game tree:

  • How close to optimal were the pieces on this specific roster at each decision node?

  • Did I give up any expected fantasy points already (did the roster have any "duds")?

  • How close were ownership projections to the actual ownership percentages for previous games?

  • How can I maximize my expected value relative to variance at each subsequent decision node?

  • Were there any low-owned pieces that are now optimal?

Skinny Stacks vs. Over-stacks

Generally speaking, both the skinny stack (a quarterback plus one pass-catcher) and an over-stack (a quarterback with three pass-catchers) are utilized at a lower frequency than their respective hit rates, meaning you are able to build intrinsic leverage without any further action. That leverage is only amplified by leaving a correlated bring-back off of the roster, which is the generally accepted way to build for game environments. That said, the data shows that a "standard DFS correlation" (quarterback + one pass-catcher + a correlated bring-back) was present on under 15% of both the optimal and Milly-winning roster over the previous two seasons, a rate much lower than the almost 30% of both optimal and winning rosters that contained a skinny stack with no correlated bring-back and an over-stack.

Theoretically speaking, the hit rate of a skinny stack would be far more prevalent the larger the slate is, while the hit rate of an over-stack would be far more prevalent the smaller the slate it (remember, we have to be closer to optimal the smaller the slate). As such, the optimal approach to this weekend's games would be to be overweight the field on skinny stack roster constructions for the full six-game slate and overweight the field on over-stacks on both the two-game Saturday slate and the three-game Sunday slate. The reasoning for the delta between hit rate and ownership rates as far as macro roster construction goes can be explained through human and crowd psychology, which we won't fully get into in this space.

The Value of a Touchdown

It doesn't take a large stretch of the imagination to envision the importance of touchdowns in our beautiful game. That said, not many realize that the value of a touchdown actually increases or decreases relative to the size of the slate. Basically, the value of a touchdown is much greater the smaller the slate and much less the larger the slate. This is the reason why backup running backs (or 1B backs in a split backfield) have exponentially higher hit rates on Showdown slates, where the value of a touchdown is at its maximum. This means that we should be looking to embrace additional variance as far as "backup" running backs go on the Saturday and Sunday slates. For this reason, consider options like Elijah Mitchell and DeeJay Dallas on the Saturday slate, Devin Singletary, James Cook, Kenyan Drake, and Jeff Wilson on the Sunday slate, and consider avoiding them on the full six-game slate.

This realization should also influence decisions made as far as one-off selection goes, as any player to score multiple touchdowns is going to be in the optimal lineup on any of the two-game, three-game, or six-game slates this weekend.

Potential Game Environment Bets

If we place the process of targeting game environments into the bucket of information known as "common knowledge" (the knowledge that we know, the field knows, and that every player also knows the rest of the players know), we can then formulate a plan to maximize expected value against that assumption. This is the same idea as placing player selection into the common knowledge bucket of information (basically, we can no longer gain a material edge simply by selecting the best players in DFS).

This is why we should see increased combinatorial ownership on the players from the Giants - Vikings game (the game with the highest game total on the full slate) and Chargers - Jaguars (the game with the second highest game total on the full slate and highest on Saturday). This is also why targeting the game environments of the other games provides a boost to expected value, as the hit rates of those games are not going to be materially lower than the highest game total games considering how close the game totals on the weekend are. Said another way, targeting game stacks from Seattle @ San Francisco, Miami @ Buffalo, Baltimore @ CIninnati, and Dallas @ Tampa Bay will carry intrinsic leverage and expected value.

The game likeliest to go most overlooked on the full six-game slate is the final game of the weekend - Cowboys @ Buccaneers. The game total is right in the middle of the pack and there is an essence of human psychology at play in that humans don't like to wait, which helps explain the notion of instant gratification that all but rules modern society. That leads to potentially my favorite play on the entire slate, the header image of this article - Mr. Tom Brady.