STRATEGY FORUM

Comments

  • tiltatout

    Hi everyone,

    I have a quick question. I am in the process of developing my NFL LU building model (a lot of work and fun!) But I am struggling to find “previous data” on which I could back-test and optimize my model. I can think of a ton of different data that are available every week that I would like to try and incorporate in some ways into the model (starting with projections from different sources, vegas odds etc…) but it seems crazy to just come up with something and not test it.

    This might seem like an odd question to you, it’s probably due to the fact that I am new to DFS and I am actually new to American sport in general (moved to NY 2 month ago from Europe ^^), so I come to DFS from a gambling background (poker, blackjack, trading…) rather than a “sports fan” one.

    How do you guys do it ? Do you just build your model on your “gut” and slowly collect data/test and adjust it ? Is there a source out there with this kind of data ?

    At the moment I mainly use FantasyPros, FFtoday and the NFL injury report because these are the only ones where you can get past data easily (and pro-football to get actual results)

    Would love to discuss with some other model builders out there ;)

    Best

  • tiltatout

    No response ? ^^ Am I fooling myself in thinking that developing at least a “semi-automated” model to DFS is the future and that a lot of people are doing it ? Or do people simply don’t want to share this kind of info ? (which I would understand but I find it a bit sad since I would’ve been happy to talk about it with some people)

  • BobGrinder

    What info are you looking for? It’s all available so I’m not sure what it is exactly you’re not finding but if you if you get specific I’ll try to point you in the right direction. You’re probably going to have to do some work getting it into the form you want though.

  • theoddsmaker

    previous data? Like passing yards and whatnot or player ownership %?

  • louiescards

    DK KoTH winner

    • Moderator

    “How do you guys do it ? Do you just build your model on your “gut” and slowly collect data/test and adjust it ? Is there a source out there with this kind of data ?”

    You find data points for each specific sport and try to find correlation to fantasy production. As for what you are looking for and how much variance is ok that is way different in each sport.

  • liquidfire

    Some people scrape websites. This is a grey area in terms of ethics but it happens a lot. Im sure most of the pros pay for their stats, as they will be easier to work with and of better quality. It gets expensive though.

    I find that getting the basic stats (yards, tds, targets, etc) is not too difficult but there are a lot of other things I’d like to build into my model like historic vegas lines that are more difficult to track down.

  • BobGrinder

    @liquidfire said...

    Some people scrape websites. This is a grey area in terms of ethics but it happens a lot. Im sure most of the pros pay for their stats, as they will be easier to work with and of better quality. It gets expensive though.

    I find that getting the basic stats (yards, tds, targets, etc) is not too difficult but there are a lot of other things I’d like to build into my model like historic vegas lines that are more difficult to track down.

    Depending on what stats you’re looking for many sites offer a downloadable CSV of stats.

    For historic lines/O/U/results here you go…from 1978 -2014.

    http://www.repole.com/sun4cast/data.html

  • yanks237

    @BobGrinder said...

    For historic lines/O/U/results here you go…from 1978 -2014.

    http://www.repole.com/sun4cast/data.html

    WOW thank you thats great info!

  • tiltatout

    Hi everyone, happy to see that I am not alone in this!

    I realize that my original message was a bit vague, and to be honest I really just wanted to start a conversation about what kind of data you can/should incorporate into a model.

    One of my main question is this: As a new DFS player/model builder I thought the first thing I would integrate into my model would be projections stats from different sources. Mainly fantasyPros/FFtodat and FoxSport because they are the only one where I managed to get past projections. (which I need to estimate their accuracy).

    Do you guys also use this as the start/ do you think it’s a good idea or do you think I would be better off trying to completely develop my own projection system (I definitely intend on improving upon other’s projection though).

    With regards to which stats exactly I am looking for, to be honest when I wrote the first post I was mainly trying to get a reliable source for past vegas odds, which I ended up finding on foxSport, a great info would be past individual player props but this looks tough to find. In general I am interested in knowing what kind of data you guys consider and how much backtest you do before including it. Currently I am studying the data over the 2013-2015 period as a kind of equilibrium between number of data point and getting data that are still relevant (the game changes).

    The other big question I have is what kind of predictive power can you reasonably achieve (in term of R^2 for example) and what are your opinions as to what the Profitability~R^2 curve looks like.

    Finally, I am interested in what you guys think is the best format to apply this model-based approach to DFS. At the moment I focus on H2H since I don’t have a variance estimation model yet, but I would guess that being able to accurately predict variance and correlation would give a huge advantage in GPPs.

    BobGrinder: Thank you very much for your link, I haven’t had time to check it out yet but I will ;) If you are willing to share the other sites you use to get historical data (whether or not it’s in a convenient format) I am sure that a lot of us would appreciate it since you seem to be particularly knowledgeable about the subject :) As to what data am I exactly looking for, to be honest I am quite willing to test anything and I have relatively few assumptions as to what may/ may not influence production and what may/ may not already be included in others projections.

    Theoddsmaker: Well, I already have the simple stats, however you’re raising a good point. Past ownership % would be a must-have to actually develop an ownership model (which would be very useful in a GPP lineup building model) Do you know if this is available? I guess it would depend on the site (also maybe not that much once we control for salary which we can probably find quite easily).

    Louiescards: I am sorry but I didn’t really understand your comment… “How much variance is OK”? At the moment I only focus on NFL, thought I would try to build one model before looking into other sports ^^

    Liquidfire: It’s interesting, I hadn’t really considered the morality issue, why would you think it is a grey area (as opposed to white ^^), at the moment my bankroll doesn’t really justify paying a lot for data so I scrape sites, however I can imagine that a pro would want to get “cleaner data” to avoid mistakes…

  • iam12fear

    http://sportsdatabase.com/ has tons of raw stats available. Once you familiarize yourself with their data query language, you can find lots of historic stats; Vegas lines included.

  • BobGrinder

    I can’t post some of the sites here on the forums. It’s against the rules to post competitor sites and if they have referral links to DFS sites they are a competitor to RG. That is why I asked what it is you’re not finding and I can PM you.

    As far as past projections or owned % I don’t know where you could get those but personally I don’t really think they would help in a projections model. I mean I see how owned % might help in building a % owned projection model but I don’t see how it would help in projecting production.

    I do know a site for DFS points scored but that could easily be calculated from stats so I don’t know if that is something you’re looking for or not.

  • DmitryPolo

    Well clearly you want an R^2 that’s as close to 1 as possible so that your data completely explains your projection. I’m sure a reasonable one would be above the 60-70% mark although I’m not sure what program you’re using. Also there’s more to it than just an R^2, like making sure your independent variables are significant and are jointly significant. Also your model could be biased so and something might be over/under specified. I’d worry more about those issues.

  • tiltatout

    Thanks for the link iam12fear, I’ll check it out ;)

    At the moment I am mostly trying to develop a variance model to try and attack GPPs, I have an R^2 of around 60% for my mean model (I of course made sure that the variables where jointly significant), I am not sure if it’s sufficient, to be honest it seems to be a very small improvement on simply using the best projections out there… (I think fantasyPros is around 55%).

    I will definitely go back to my mean model, but I think that there might be more value to a “fully automated model” in GPPs and I would like to get on it as soon as possible :)

    Bob, for past owning%, it was to build an own% model, it would be useless in a projection model since we don’t really have the info at the time… However, I could actually see the own% projection have a non-negligible influence on the production (“advanced wisdom of the crowd”).

    DmitryPolo: I was taking those things for granted, let me rephrase the question: Provided that we have an unbiased model, using significant and jointly significant variables, what kind of R^2 do you think you need to be a) Break Even, b) 10% Roi ? (Of course I don’t expect you to answer a precise number, I am merely curious as to what you may think) :)

    With my very little experience I would say that 50-60% would probably break even in low-stakes cash-game with a little “player selection” and 75-85% would crush everything up to the biggest stake… (At least it seems so because it looks pretty frickin hard to get at 75% ^^)

  • X Unread Thread
  • X Thread with New Replies*
  • *Jumps to your first unread reply

Use our links to sign up and deposit on sites listed in this thread to get these bonuses:

  • FanDuel

    Get 1-month of RotoGrinders Premium for FREE (a ~$40value) by signing up through one of our links!

    Learn More
  • DraftKings

    Sign up for DraftKings using a RotoGrinders link & receive our DraftKings Premium content FREE for 1 month. That’s a ~$40 value! No DraftKings promo code necessary!

    Learn More
  • FantasyDraft

    FantasyDraft strives to put players first, with a mission to “provide a fun and fair experience for all.” To this end, the site has a well-built, easy-to-use interface and a the first of its kind in offering “Rake-Free” fantasy contests.

    Learn More

Subforum Index

RotoGrinders.com is the home of the daily fantasy sports community. Our content, rankings, member blogs, promotions and forum discussion all cater to the players that like to create a new fantasy team every day of the week. Our goal is to help all of our members make more money playing daily fantasy sports!

Bet with your head, not over it!
Gambling Problem? Call 1-800-Gambler