Can anything stop the rise of daily fantasy games?
A massive cheating scandal, calls from Congress for more scrutiny and, now, action by the state of Nevada hasn't made a dent, as the business at the industry leaders, FanDuel and DraftKings, continues to grow.
So, if you're going to play it, why not play to win? And, since the only way to reliably win is to be a math genius with huge resources or to cheat (as demonstrated by DraftKings employee Ethan Haskell, who made $350,000 on information he acquired legally performing his job within the rules of his employer at rival site, FanDuel), why not cheat?
Aside from that it's immoral, it's also hard. Figuring out the math behind how it works requires more facility than most people have. Luckily, you don't have to know what Bayes' theorem is to know how to use it. In fact, a research firm called DataScience recently sent me a really nice breakdown of how it works. That, below, with my comments:
Bayes and Weekly Fantasy
Knowledge of an athlete's popularity for other bettors can dramatically increase our chances of winning at weekly fantasy football.
Our hypothesis might be, jokingly, envisioned to result in this graph:
TheStreet: Obviously, this means you need to have the data of how popular athletes are for players to choose. Like Haskell, you have to have access for this method to work. While I wouldn't suggest you do anything illegal, there are clearly ways to obtain that data if someone wanted to, either by hacking into DraftKings or FanDuel (is anything unhackable these days?), or by simply knowing someone who works there who might be inclined to give you the data, perhaps if you offered them a share of the winnings.
In the past, employees at DraftKings and FanDuel could use the data to their own advantage, as Haskell did. The companies have since banned any of their employees from playing the games.
It should also be said that just knowing what other players are popular isn't all that's required. You also have to have some idea how well each player might do. Because of the huge numbers of people who play fantasy sports (and I'm among them, though I have never played daily fantasy; I have been in casual, season-long fantasy football leagues with my friends for years), there is a huge industry that has arisen to solve this problem.
It even has its own celebrities, like Matthew Berry and Eric Karabell, both of ESPN. The leading sports cable network, which doesn't spend much time discussing other forms of sports betting, lavishes a tremendous amount of attention on the fantasy football implications of player aptitude, team performance and coaching decisions.
In the past, those factors were considered more for how they impacted the sport and the play; those days are long over. By the time game-time rolls around, fantasy sports players have a pretty good idea of how players might perform.
Back to the DataScience memo:
Let's assume we choose Brady and he scores 5 fantasy points more than the average of all the other quarterbacks. If everyone chooses Brady, then the advantage we have over other bettors is zero.
However, if nobody else chose Brady, than we have an advantage of 5 fantasy points more than every other bettor.
If 80% of the bettors choose Brady, than we have an advantage of 5 points over the remaining 20% of bettors.
So a simplified equation might be:
Gain from Pick = (Points Advantage of Player vs Other Players) * (% Bettors Choosing Other Players)
So if Brady scores 5 points better than other quarterbacks, and 80% of bettors choose Brady, our gain from the pick would be Gain from picking Brady = 5 * 20% = 1
Now, let's say Andy Dalton's fantasy score was better by 2 points than the average of all the other quarterbacks.
And, let's say, only 20% of bettors chose Andy Dalton, so our advantage is over the remaining 80% of bettors who did not.
Then we'd have:
Gain from picking Dalton = 2 * 80% = 1.6
Dalton beats Brady!
TheStreet: As many football fans (and even season-long fantasy football players know), in no other universe does Dalton beat Brady:
-- Quarterback rating in 2015 (through the first five weeks of the season, Brady: 121.5; Dalton: 115.6)
-- Career win percentage (Brady: 78%; Dalton: 63%)
-- Super Bowl wins (Brady: 4, including 3 where he was selected as MVP; Dalton: 0)
This kind of math -- and having the knowledge necessary to apply it -- is what allows someone like Haskell to waltz to victory by being "smart" enough to pick a seemingly random assemblage of players the average competitor might not expect.
In its advertising, DraftKings and FanDuel tout that it's the bettor who believes in the player that nobody else believed in who wins. In a way, that's true -- it's just that some people have a much better reason to believe than you might: they know nobody believes in them because they haven't chosen them and that they have a pretty good chance of performing well.
Back to the DataScience memo:
Dalton won because he is picked less often, despite having scored less than Brady.
This "rarity" score increases his value to us.
So, if we know how many bettors are likely to pick each athlete, then we have insight into the rarity score.
Rarity in a Portfolio
Haskell's pick on Fan Duel are public, as are the distributions of ownership after each week. We can see from the chart below, made from the Fan Duel data, how rare each of Ethan Haskel's picks were, as well as how well they scored.
The cluster in the upper right corner that includes Olsen, Cobb, Freeman, Green and Dalton feature a particularly powerful blend of rarity and high scoring. Although projected points are not guaranteed, if the bettor knew the rarity of players, these five would've been excellent choices that leveraged that rarity.
TheStreet: This knowledge is made more powerful by people or organizations that can afford to create many, many lineups on which to bet. While the combination above worked last week, there were many equally exotic combinations that Haskel and others who had prior knowledge could have created that would have looked good on paper, too. It's just that, often, the players don't always perform as expected. As they say in the real sports world, that's why we play the games.
Back to the memo:
Including Prior Knowledge: Bayesian Inference
Quantifying exactly how much knowledge of ownership distributions would help a bettor requires modeling odds of points and salary cost along with this prior distribution. The practice of including prior certainties of specific factors in an equation is known as Bayesian statistical inference, a technique frequently used in the machine learning and statistical analyses we work on here. We use these techniques when creating statistical models such as customer lifetime value or market predication as a way to better guide decisions by quantifying our certainty.
As a way to illustrate the value of Bayesian inference, we may build out a more detailed model of the impact of prior knowledge of ownership distributions on weekly fantasy sports odds, and we'll be sure and post an update here when we do.
TheStreet: People don't play fantasy football because they think it's a meritocracy and they have a real shot of winning. Let's face it: People like to gamble, and that's what this is. We know the casino is rigged in favor of the house, yet people still go. Some happily spend their lives pulling the slot machine lever. The key point here for the future of daily fantasy is, once regulators figure out that DraftKings and FanDuel are more like casino gambling than not, will government take action?
If I was a betting man, and I am, I would say so.