Hey guys,

I'm late to the DFS party and I've only recently begun giving it any serious thought. FWIW I'm focusing on DK's NHL contests...

Thinking of tournament poker's ICM calcs, couldn't every player on a given card be reduced to one equity-like number in the same vein as a poker hand?

e.g. Player-A (three-game card):

*Salary: $6,500
*Expected FPG (adj. for tonight's competition/venue): 3.15
*Projected Own-%: 25%

Like our opp. range in poker, own-%s are the unknown. The more accurate a player ranges his opp., the more accurate his equity calcs.

It seems the first step would be quantifying the effects of own-%s on a player's value. i.e. Player-A (@ $6,500 and 3.15) is worth x @ 15% owned and y @ 30% owned. Clearly, own-%s cannot be viewed in a vacuum--30% on a three-game card is not the same as 30% on a 10-game card.

I'd love to hear any/all thoughts regarding how (if at all) I might quantify own-% effects as well as any other bright ideas you might have.

I appreciate it--thank you.

you don't seem to understand what icm means

There are basically three ways to approach it: (a) analytically, (b) simulation-based, and (c) empirically.

(a) An analytical model is an actual proof with a closed-form solution. You have to make a bunch of assumptions about scoring distributions, the joint distributions between players, the payout structure, etc. Then you can backsolve for optimal strategies either by hand or with a computer. You probably shouldn't try this if you aren't well-versed in theoretical statistics / mathematics.

(b) Simulation is likely your best bet. I'd start with a really simple model like a 3-man contest with a simplified game format, e.g., everyone drafts 1 player. Set the strategies, then run the contest a million times and see what the equities are. When you are satisfied with that, add one level of complexity (e.g., draft 2 players) and see what changes.

(c) You could download a bunch of DraftKings CSV files and try to estimate it empirically, but I wouldn't recommend this approach because I'm not convinced it would produce anything useful.

You're gonna need some coding chops to pull off (b) or (c).

I did (b) for NBA. In the first step i created a tournament with 10 000 different teams all within a salary near the max and with real ownership of the players.
Than in the second step my program simulated 10000 different outcomes using a gaussian distribution of projected points for each player.
Third step i looked which teams were in the top 1% of the field most often. It happened that ownership % was completely irrelevant. Teams with best value players perfomed best, even if their ownership was high.
But iam not a programer and my model has flaws. In reallife there might be more outliers in score which could favour low ownership. Also 10k simulations are not enough but the program was so slow lol.

Thank you for the thoughtful replies, guys. I came across this article on a Google dumpster dive. Any further thoughts would be appreciated.

https://www.4for4.com/fantasy-footba...-ownership-dfs

Suggestions

1: Don't use 10,000 teams. Start with a 100-man or something.

2: Examine extreme cases. For instance, set up the scoring so that one player is such good value that 99/100 "rational" entrants will roster that player. Give the 100th entrant a strategy that strictly fades that player. The scoring will not follow a normal distribution, but more likely a negative binomial distribution that is itself the sum of multiple negative binomial distributions.

There is no proof here for why this method would be best. It just seems completely arbitrary to me. Maybe it's right and maybe it's not, but there's really no theoretical or empirical justification for it.