Those of you who follow me on Twitter will have seen that I had some great results in the NBA over the past week or so, as I managed to follow up Draftstars a win in $30k win last Thursday with a number of strong lineups on Friday, and another win today that has made it’s way into this post:
With those three tournaments, and the rest of the weeks results, my total prize money on Draftstars for the week came in at $13,284. Add that to about $1800 in prize money at Moneyball, and the total profit for the week was well in excess of $12,000 after entry fees!
But, you’re not really here to hear me talk about my winnings, or how important a tool like the lineup cruncher is when you’re entering a number of lineups each day! Those wins prompted a number of questions on Twitter, Facebook and E-Mail that I answered, and we thought it might be interesting to share some of those with you.
Q: How has the NBA season been for you?
I get this one quite a lot. We have a reasonable number of people using our tools now, and I often see their names on the leaderboards. But, of course not everyone wins every day.
Overall, the NBA season was (until this week) a mixed bag, I had been running at around 15-20% ROI across the two platforms, but that was made of probably less than 7 great days, combined with a whole bunch of average to bad ones where I made my money back or lost a bit. And, I should caveat this by saying I’ve been entering 6-15 lineups per day, per platform, on probably about 80% of the days this season.
The challenge with NBA is that so many of the players score so closely to each other that you need close to the perfect combination on any given day to even be in the money.
Q: Any tricks of the trade you can share?
One thing I’ve been doing recently that helps is looking at the combinations of players that I’m including.. e.g. with the Warriors, Green and Curry make a pretty good combo – when one does well the other tends to, while Durant and Curry seem to take points off each other, so it’s generally a bad idea to have them both in the same lineup.
I’ve actually been working on an article, a table (and ultimately a quick tool) that lets you map this for lineups you’re considering building, by showing the (statistical term warning!) correlation and covariance between two players. No promises on the timeline for this, but aiming for early next year.
Q: Do you lock in certain players or just narrow down each position than let lineup cruncher do its thing ?
Generally, I use a mix of 3 approaches on a day-to-day basis:
i) lock in 3-4 players that have good $/pt, projection and upside potential, and build the rest of the team around them using the cruncher to generate options.
ii) set 2-3 player stacks for a team; paying attention to which players have a positive correlation/covariance (see above), and then using the cruncher to generate the remainder of the lineups.
iii) pick 4-6 players at each position and build from that. At the moment, to do this on the current lineup cruncher means that you have to go through and manually excluding the rest of the players. However, our new lineup cruncher will have a shortlisting functionality designed to make this much easier!
Having a play with the new lineup cruncher before it comes to you guys very very soon pic.twitter.com/YZXIqNOgQu
— Fantasy Insider (@FantasyInsider) December 13, 2016
Q: How do you get different teams to your subscribers if you just use your model?
We say this quite a bit, but we don’t recommend anybody just enters the lineups straight from the cruncher without changing any settings. I know some users do this, and they do it profitably, but the expected value (i.e. long term profit) decreases with each new person doing it. You will be much more profitable changing one player, even if that lineup makes your team 0.5pts worse on the cruncher. You might be 0.1% or so less likely to win (but probably not even that), but winning the Draftstars $30k vs tying for first is at least a $750 difference. That 0.1% less chance of winning is maybe costing you 0.1% of $1500, or $1.50 of equity in the $30k contest. In the long term, unique lineups are better.
By using the approaches described in the previous question, I pretty much always end up with a unique set of combinations.
Q: Is your background in computer programming/engineering? Do you watch all the games?
Katie & I watched a Boston Celtics game in the US last year while we were over there on a work trip. That was the first full game of (professional) basketball I’d ever watched – though I do watch quite a bit of the March Madness tournament. The last quarter of the GSW game as I was hanging on to $30k was the first quarter of basketball I’d watched this year. So, no, I don’t really watch a lot. My opinions, team selections etc are all quantitative – everything I need to know is in the data somewhere (though data here also means reading social media, scanning US fantasy sites etc).
On background, mine is pretty mixed. I did some Computer Science and Engineering in my Bachelors and Masters degrees, but I did a humanities PHD. I then worked in data science for a few years before I left academia.
Most of the stuff I do now, and here in particular though is self-taught; I built models for US Sports from 2002-2010, and gambled semi-professionally alongside academia using them before the UIGEA passed. I took a few years largely off to focus on getting the PhD and looking at an academic career, but got the itch again working on the social media datasets, and here we are. We now have pretty solid models for Australian sports, and you’ll see a bunch of those in the next 12 months both here and on Stats Insider.
Q: What sort of info do you look at to make the projections?
The projections use machine learning to build a distribution of player performance in a range of metrics, and then a Monte Carlo / Markov Chain (depending on sport) approach to ‘simulating’ each match 1000-100,000 times. At the end of that, we have a distribution of the players fantasy scores.
We feed in to the system a number of factors, including the players historic stats, but the three biggest factors for NBA are estimated playing time, some custom analytics based on pace, and what kind of statistics an opposing team gives up historically to ‘Players Like Draymond Green’ (i.e. those players who have a similar statistical profile to DG).
Q: Do your programs do projections for the season long games like Supercoach/Dream Team etc? Do you play those games as well as DFS?
AFL Fantasy/RDT yes, Supercoach no. As far as I’m aware Supercoach scoring system is still something of a secret / not possible to replicate with public stats. I do often have a team, but often lose interest fairly quickly in non-cash games.
That said, we are looking at some exciting crossovers this year that may impact on what kind of season-long content you see on the site, so stay tuned!
Q: When you set your line-ups do you have a ‘favourite’ line-up and how do they usually go compared to your others?
I really don’t.. I work out how many entries I want (roughly 1 per $1k of prize money in NBA), split those into ‘sets’ of 3/4 and work out a logical construction for each set (e.g. stacking GSW, focusing on key players/teams etc), then use the lineup cruncher to build that set and enter them. It normally takes me 30-40 minutes per day.
Q: Is it a good idea to stack teams in the NBA?
It’s an interesting one; in general in GPP’s I think some stacking is useful, but it’s not quite as simple as in NFL (where you match QB with WR or TE) or AFL (where players tend to have a strong correlation to the team’s final points). I ran the data on this recently and there are definite patterns within teams play, which I touched on in the correlation/covariance answer above.
Q: What are your recommended setting for NBA lineup cruncher for GPP’s ?
For NBA I tend to use the defaults; we’re tweaking the sliders a bit in the new version of the cruncher but with NBA I think the biggest single key to accurate projections is in playing time, which are factored into the regular projections. If you focus too much on upside or floor, you can end up weakening the playing time effect on projections too much.
If you want to tweak to make unique lineups – which I sometimes do – I normally assign 8 to the default projection and 2 to consistency for double ups, and 8 to default and 2 to upside for Guaranteed tournaments.
I hope some of this has been helpful and feel free to ask your own questions in the comments below!
We’re getting ready to head into a busy stretch, with big NBA tournaments over the next couple of days, and $10k+ tournaments coming up for Big Bash starting next week. We’ll have more on our Big Bash package soon. Also be sure to take a look at Stats Insider, and sign up join the waiting list for free early access when it’s up.