Hacker News Reader

Comment by zem | Hacker News Reader

someone please explain like i'm five - according to the wikipedia page, fantasy football scoring is based on the actual performance of the athletes you pick in an actual game. therefore, if the games themselves are not fixed, there should be no possibility of "insider information" - what information could the companies running the pool have that the general public do not? the article didn't make it any clearer.

View on HN

The advantage is in knowing what percentage of football players are started by the entire pool entrants. This information is released once that particular game is started (e.g. % start for NY Giants players would be known to all at 1 PM when their game started and % start for Denver Broncos players would be released at 4PM when that game started), but was known ahead of time by those who worked for the company. While these %'s would not be exactly the same between Fan Duel and Draft Kings, they should be close enough.

How does this all help you win? You can get an average return in a tournament (e.g. buy in for $10 and win $15 or $20) by playing high point, low variance players (safe bets), but to win big, you need to play high variance players (e.g. the wide receiver who will probably score 4 points, but might score 20-30 pts if they have their best week of the year). That said, if everyone plays the same high variance player, you will be beat by or end up tying with many other players. So, if I have two high variance players that I think may have a good week, one who is owned by 8% of teams and one who is owned by 1%, I am better off playing the 1% player who will put me ahead of the field if he scores big.

Edit to add information for those that aren't familiar with daily fantasy:

Each player gets a salary and you are working under a salary cap that prevents you from making a team consisting entirely of "studs" (the top players at each position). If salary cap is 10X, then the average salary on your team needs to be X. A "safe bet" like Tom Brady or Aaron Rodgers will cost you close to 2X. So for every safe bet you need to pick a long shot like Tavon Austin (who Yahoo tells me was owned by 1.4% of daily fantasy players) with a salary just over X/2.

But if you follow that strategy you might have a +EV but you are very very likely to lose in any given entry because you only win when all (or most of) your high variance players come through. And that is, by definition, unlikely.

So this guy won big the first time he tried this? Or maybe he's tried 4 weeks now and has already won big?

This still doesn't add up.

Undoubtedly the guy has a system and has diversified in either time or by parallelizing in several "leagues."

You can enter multiple times I believe with different lineups. So you enter some lineups with high variance low % picked players.

But if your system is to pick high variance players that are also underrepresented on other players rosters how many lineups can there be? I remain skeptical that this is a viable strategy.

Here's an experiment: Go enter some giant free-roll and a $10 tournament with a huge guaranteed prize pool. I would bet that the top score (not the average) in the free-roll will have higher score. Now take the points that each player scored and their original cost and you can use a variant of the knapsack problem[1] to figure out the optimal lineup. You'll see that quite often the optimal lineup wasn't picked by anybody (mostly because the player costs are pretty good and the optimal lineup sometimes doesn't use the full salary).

[1] https://en.wikipedia.org/wiki/Knapsack_problem

Sure, but how would access to internal data help a DraftKings employee pick that optimal lineup ahead of time?

To clarify, you don't just pick high variance players.

You _must_ pick some people that are high-cost high-reward types. The superstars that will almost always get their lion's share of points. You cannot win a tournament just by picking high variance players, or it would be extremely hard to do so. You also have a salary cap so you can't just pick 5 superstar players. The higher their perceived worth, the higher they cost to "buy" them.

So the main point you can differentiate yourself from other players is to find "sleeper" picks. These are players that are OK but not GREAT but if given a proper matchup, could do very well.

For example, you may pick the 2nd best receiver on a team hoping he has a great day because the opposing defense is known to double team the offense's best receiver. This may give the 2nd best receiver more opportunities for catching the ball for yardage/touchdowns.

BUT the thing is, this happens a lot. So you know maybe 4 games where this will happen. But you don't know the percentage picked of each player. Just like in March Madness brackets where picking all chalk (the favorites) to win is a bad strategy because almost everyone else is doing it and at best/worst you will tie, it is to your advantage to bet on players that you think _may_ have a chance, but are picked the least from the pool.

So the advantage this employee got, was he knew how often each player was picked on DraftKings, so given his basic assumptions he just applied those percentages to FanDuel. He knew player X had a good matchup, but _no_ one was picking him. So he plays multiple lineups with player X, buying him up so if he has a good day the employee will win big.

But he also places wagers on player Y, player Y has been picked more than player X, but still enough where the employee can get a good return, so he places a couple of lineups with player Y in it (even combining with player X hoping they both have big games). Buying some lineups with player Y also hedges your bet in case player X doesn't do well. This is probably what you are curious about. You don't pick only player X hoping he does have a great game. You hedge by buying more lineups with him in it, but supplement that with other lineups and picks that you think are +EV.

So the advantage is you know who hasn't been picked a lot. And as you start buying players you have to start spending more "efficiently." If you know player Z has only been bought 10% of the time but you know he has a high likelihood of a big game, you will pick player Z more in your lineups than other lower priced players because his returns will be good.

I agree with all of this and I agree that one could probably construct a +EV strategy this way. I'm just surprised that the strategy could pay off so shortly. I would imagine that you'd get a scenario where you had a 1% chance of winning but a payoff of > 100 to 1. So +EV but still likely to take a while to come to fruition.

Top fantasy players make hundreds of submissions per tournament to smooth out variance.

that edit is very helpful for me, I've recently starting playing fantasy e-sports and I didn't fully understand the proper strategy for salary cap pools.

Let's say that you, zem, are an expert player and you have entered a roster into DraftKings Million Dollar contest.

I, swanson, am an employee of DraftKings and have access to your lineup and/or data about all the other lineups entered. (* This is speculation)

I, swanson, then use this inside knowledge to make my own lineup on FanDuel (I am not allowed to play on DraftKings because I work there). I win big money on FanDuel.

---

The results of the NFL games are not the inside information, but rather the game theory-esque information about the decisions made by other parties in a zero-sum game. Knowing which players the elite top 1% players are playing is an advantage. Knowing which players appear in the most (or least) lineups is an advantage. This data is only available to insiders (employees of the companies), so that is where the "insider information" analogy is coming from.

This is the point that needs to be made, that most people seem to be missing, including tfa. Well said.

This is the only explanation that makes sense to me. All explanations where that are saying knowledge of ownership %ages in general gives some sort of edge don't really add up.

From the article:

The data that DraftKings acknowledged was released by its employee, Ethan Haskell, showed what particular players were most used in all lineups submitted to the site’s Millionaire Maker contests. Usually, that data is not released until the lineups for all games are finalized. Getting it early, however, is of great advantage to make tactical decisions, especially when your opponents do not have the information at all.

And to translate that to a potential game strategy, in large contests like these you generally want to zig when everyone else zags. Think of it like a lottery where you know what numbers everyone is playing beforehand. If 100 people bet 12345 and 0 bet 54321, you are better off choosing 54321. It won't necessarily get you a win, but it will help increase you expected value of return by knowing which popular numbers/players it is smart to avoid using.

Even with that, you'd still have to be good enough to pick the best roster. I don't think it's as useful as the article makes it sound.

Not true. If anything, it's even more useful than the article makes it sound. If you knew absolutely nothing about sports, but had access to this data and a fairly basic understanding of probability and statistics, you could use this data to be basically +EV in any large field contest you enter. The lottery analogy by slg is a great metaphor to understand it.

This Deadspin article does a solid job explaining it: http://deadspin.com/draftkings-employee-with-access-to-insid...

Another very illuminating read: https://rotogrinders.com/threads/draftkings-ownership-leak-8...

Huh? You just use the company's internal pricing model and use it on other sites once you know discrepancies. It's not "simple" but it's not genius level derivatives trading, either.

What? I can see the prices of players on DraftKings and other sites as a player. That "internal pricing model" is public knowledge.

The controversy isn't about knowing player price points. That's like the gameboard and it's known to everyone. It's about knowing ownership percentages - which is the overall percentage of lineups in a contest that drafted a certain player.

That info gets published once the contest starts (it's very compelling data), but having access to that ahead of time is the equivalent of insider trading - because you have knowledge other people don't have when they make their picks, which provides you a completely unfair advantage.

Yes, I realize that player owned percentages is a huge advantage (to bet against the field thus having a better chance of coming up). But not player pricing.

Oh.. right. I think we're saying the same thing then ;)

Those are the advertised pricing for players. If you do not believe that these companies have a private, internal model to set those "prices" to maximize their returns...

Their return is constant and only dependent on the number of players. They have no need for "secret prices"

No, but the parent comment is not that far off. Their prices are public but with the weight and velocity of betting coming in, it informs them what the lines probably should be instead.

You are correct. My comment is wrong.

Except Draft Kings lets you enter 1,000 games per week so one can use portfolio theory with this information to increase the expected value of playing.

so it's basically using the "wisdom of the crowds" that play on the DraftKings site to put together a really good pick? that still seems like it shouldn't count as equivalent to insider information, because it's aggregating a bunch of guesses that have no information either.

No. If 60% of people pick Player X, his value goes down. This is parimutuel betting. You don't want significant overlap if you can help it; how would you beat the crowd otherwise?

You find outliers and apply a pricing model that finds the best gaps. The information is incredibly valuable. Take it from someone who did this for a living at one point.

This isn't quite parimutuel, because in that system all the bet totals are published in nearly-real-time. If you wait until 1 MTP to place your bet you have a very good idea what the odds will be. It isn't obvious to me why the sports gambling sites need to keep this information secret, other than to create exploitable situations like those described in TFA.

They don't need to keep the information secret. They could operate just like pool betting if they wanted to. The problem comes when the information is kept private but insiders can see it. This would be equally bad for pari-mutuel betting.

aha, looking up parimutuel betting made it a lot clearer. thanks!

No problem. It's intentionally hard to figure out. Parimutuel wagering is inherently extremely confusing on purpose.

"The data that DraftKings acknowledged was released by its employee, Ethan Haskell, showed what particular players were most used in all lineups submitted to the site’s Millionaire Maker contests. Usually, that data is not released until the lineups for all games are finalized. Getting it early, however, is of great advantage to make tactical decisions, especially when your opponents do not have the information at all. A spokeswoman for DraftKings said Haskell simply made a mistake and that the company was certain that he did not use the information improperly. She declined to go into specifics about the safeguards or the company’s auditing policies."

I'm guessing that Haskell assumed that if 60% of people on DraftKings were going to pick [Fantasy Player X], then about 60% of people on FanDuel would pick [Fantasy Player X], and knowing so ahead of time informs the odds on each bet?

This reminds me of the Capital One/Chipotle incident (http://www.bloombergview.com/articles/2015-01-23/capital-one...).

so he was trying to bet on which way the odds would be placed and get in ahead of that, with some sort of hedging that would work regardless of how the games turned out? i have a feeling there's some key insight i'm missing here - is this like placing a bet on a dark horse before the public catches on and the odds swing closer to even?

That's what my understanding is, although there's a lot more informed analysis in this thread now.

In the late 90s, I knew a guy who ran a free "pick the winner" football web site. He said he did it because of the phrase, "the masses are the asses", which means that the general public as a whole bets incorrectly and loses a majority of the time.

This may be a bit of a simplification of what he actually did, but it was my understanding that he would see the trends in how his users were betting for the week, then looked at at past performance and used that info to make bets himself. He started making quite a bit of money doing just that.

So generally, if you have data on how people are expecting games to go and have knowledge about the different betting and/or gaming sites, you have an advantage over the other players.

I'm not sure this still holds true with the advent of betting exchanges. The market is now pretty efficient. Sure, there's still value to be exploited if you're good enough. However, in general the wisdom of the crowd is quite close to the true odds, at least at the short end of the market.

Efficiency in Exchange Betting Markets

http://faculty.london.edu/mottaviani/exchange-betting.pdf

"In terms of the Sobel and Raines model, a lower conventional bias in the exchanges relative to traditional betting markets is consistent with a higher proportion of ‘serious’ bettors on the exchanges than with bookmakers"

Efficient Market Hypothesis

http://www.betfairprotrader.co.uk/2010/12/efficient-market-h...

"If you were to take all horses starting a race at 2.76 then you would find their win rate was pretty close to 36%"

Market Efficiency: An Analysis of the Internet Betting Exchange Market

http://studenttheses.cbs.dk/bitstream/handle/10417/2888/fred...

I think when assembling a team on these websites that you get some fixed amount of cash to buy players to put on your team. The website uses the data on how many people play each football player to generate the cost of these players. I think this means then that employees can use this understanding of the value of players and the relative frequency that they are chosen to pick smarter.

I think in general, one would benefit from accurately picking players with great expected values who are also generally under-picked.

Reply on news.ycombinator.com