No Frills DFS Data - Arnold Palmer Recap & Reflections on Exposure
This slate was a really nice breakthrough for me, not so much in results but more in further confirming a few things for me.
First of all, I had a 4/10 cash rate in gpps but all basically the min cash variant, which would have been bad if not for the fact that 1 of those 4 was also the lineup I chose for single entries and another the one I chose for cash. So where the majority of the money was played, the lineups did very well for an overall 16% ROI on what could have easily been a losing slate.
Thoughts on Cash Lineups
Secondly, my cash lineup was one pivot off one of the nosebleed players and his clique. I unfortunately had the bad end of that pivot but was still a pretty good lineup and managed to grab one of the last remaining pay slots behind that train most of the time. However, I see a lot of positives in that. One of my earlier struggles was figuring out which lineups were suitable for cash as no single algorithm seemed to be most reliable. So lately I’ve been selecting by hand and the results have been pretty good. In fact, this is the 2nd week in a row I nearly mirrored a few of the top players cash lineups so that’s given me tons of confidence.
For what it’s worth – the moment when I decided to stop being extremely cautious and really get into cash games was when I noticed that Moklovin, Hoop and some others nosebleed players and I were usually only a few pivots away in our cash lineups (and more importantly their pivots were ones I considered myself). The second realization was when I realized that none of the sharps were sitting my h2hs voluntarily anymore. That’s truly when I made my breakthrough. You gotta feel at least a little bit of pride when people who sit 10k h2hs every night won’t bother playing you for $1 anymore because they don’t want to trade that 10 cent rake each night. That’s when I decided to step it up and stop limiting myself to $1 and $2 h2hs. I correspondingly have begun offering higher stakes h2h games this upcoming golf slate than I normally have done in the past. I won’t go as high as in other sports because the variance in golf is massive but I feel I’ve been playing a bit too conservative and will open it up a little wider.
Beware the 10 man trap
I’m also more concerned than ever about playing 10 man groups. Simply put, the sharps do so much volume that they nearly always constitute half or more of those groups. Meaning that it’s far more difficult to cash in those than larger groups or even h2h (where you’ll face a more random population). I cashed in only 72% of my 10 man 50/50s but 100% in anything larger than 20 people. The cash line in 10 man 50/50s I played had as much as a 50 point differential depending on how many sharps sat it. I also won 89% of my h2h for reference. Had my one pivot off the syndicate train been better than the syndicate it would have been closer to 100% cashing in 10 mans, but I can’t really complain, honestly I welcome them putting 2-3 guys in a 10 50/50 and if anything, it makes me think some of them just may have come up with that lineup naturally like I came up with mine. I know 3 of them work together because those 3 always have the same cash lineup in any sport and of those 3 only 2 were ever together in a single 10 man – which seems reasonable rather than sharing 3-4 lineups in a 5 payout cash game. The others could just be random as the cash plays were all very chalky. With exception of Poulter all my cash picks were over 50% owned in most groups. You may guess then that Poulter was my pivot off the train. Had my model been forced to spend max salary then I’d likely be all aboard that train myself.
In NFL I would carefully scrutinize each 10 man and if there were as many sharps present as payout slots I’d leave it alone which meant I maybe played in 10% of them. In basketball I don’t even bother checking because I know what’s there already and in hockey I often don’t bother checking unless it’s a weird slate as some of the better players prefer to sit those out – I did too until I realized that the guys I was most worried about usually weren’t present. Baseball is my best sport, so I imagine I’ll go hard in 10 mans at the beginning of the season and play it by ear. For the record, I have a slightly positive ROI in 10 mans but without careful game selection I’m sure it’d be deep in the negative. In golf, I’d previously found the 10 mans to be much weaker than the 10 mans in other sports, but now I’m beginning to think maybe I wasn’t paying enough attention as this was the first time I played some high dollar ones so I paid more attention and then that led to me further scrutinizing the cheaper ones too. Will definitely take a stronger analysis of who exactly is in the 10 man before I commit to it going forward.
I really think sites should do something about limiting the amount of 10 mans players can play. I’d also like to see them do what many poker sites do and forbid too many associated accounts from sitting the same table together. I’m not saying we should label them as colluders because like I said earlier, my lineups are often very similar to guys like Hoop and Moklovin, I imagine plenty of people saw me with same lineup as them and thought I work with them too – but there are ways to put in policies to ensure that people don’t feel uncomfortable about it. The sites should do more like poker and treat the tables like Caesar’s wife. When I lived in Macau it was an annoyance that Pokerstars wouldn’t let me sit in on certain cash games or sngs because there was already enough players with Chinese IPs at the table, but I understood the policy and frankly agreed with it. I’d rather be forced to click another button than myself be sitting at a table where it could be 3 or more guys sharing information against me.
The Player Pool
Enough of a rant, here’s a recap of my player pool
T10 Lucas Glover
T6 Rory McIlroy
T63 Justin Rose
T10 Jason Kokrak
CUT Michael Thompson
CUT Cameron Champ
T40 Rickie Fowler
T17 Chesson Hadley
T17 Henrik Stenson
T33 Hideki Matsuyama
T23 Ian Poulter
T58 Steve Stricker
CUT Brian Gay
CUT Brooks Koepka
T46 Bryson DeChambeau
CUT Corey Conners
CUT Daniel Berger
1 Francesco Molinari
WD Jason Day
T46 Joaquin Niemann
T46 Keegan Bradley
T10 Luke List
T33 Ryan Moore
CUT Stewart Cink
T3 Sungjae Im
Players the model selected but I didn’t go with
T17 Bubba Watson
T10 Byeong Hun An
CUT Danny Willett
T6 Keith Mitchell
CUT Louis Oosthuizen
T6 Matt Wallace
T50 Patrick Reed
T3 Tommy Fleetwood
Not going to kid myself, from all appearances I didn’t do the best job in lineup selection. While the chosen core did perform pretty well, of those 8 I didn’t play, 4 were T10 or better. So I took a look at what could have been.
The Player Pool ROI
Now the lineups I did play ended up in a 16% ROI but as stated, that’s due to disproportionate volume on the two lineups which happened to do well. So running the numbers, if I played them for the same amount each, then I would have had -10% ROI in cash and -8% ROI in gpp.
So let’s look at how the other lineups I didn’t player would have fared. This is where it gets pretty ugly, while I didn’t bench anything that would have gotten top 10 or anything, I did bench my best lineups. They would have had a 70% ROI in GPP and 26% ROI in cash. That’s pretty damn ugly. One lineup in particular would have had a legit shot at winning it all as it already had a top 50 finish despite having Cameron Champ and his +9 disaster.
All added up, I would have been significantly better off selecting other lineups, but the silver lining is that of those lemons I went with I managed to pick 2 good ones where it counted. Furthermore, while breaking down the slate I discovered that my single worst performing lineup was my exposure control lineup, my 2nd to worst lineup was my homer lineup. While historically these lineups have fared alright, I’m going to still make them but let them ride the bench this next upcoming slate and put a bit more faith in my model. Twice now the guy I wanted more of and brute forced in (Adam Scott & Cameron Champ) have not performed well at all. Despite that this is a negative result, I actually view it positively. I’m not ready to quit Champ yet, but having taken a look at the upcoming slate, I would be surprised if he’s played this week. I’m assuming my model is going to push a lot of Spieth, Koepka, Kokrak, Mullinax, Garcia & Finau but I’m often wrong about my early predictions just from seeing prices.
Another thing I noticed is I accidentally played the OWGR model by accident, I don’t know how it happened, but the data for that lineup was copied over twice and ended up replacing one of my more more reliable models that is an automatic play. Had I actually played the model I’d intended to play it’d be a 4x payout instead of a 0. For the life of me, I can’t find any bugs in the code that would have allowed that to happen. So I am assuming I must have done it myself but can’t figure out how or why. The scarier thought is that I have a lot of bugs that I’m personally not equipped to notice – not surprising given that I’m basically learning this as I go. Very sketched out by this happening and just goes to show I can never be spending too little time confirming the results. That’s something I could have caught in a minute had I bothered to look. Instead I was too concerned over my massive exposure to Rory.
Thoughts on Exposure
Speaking of exposure. I’ve decided that going forward, so long as it’s not a punt play, I’m not going to worry so much. Glover was so undervalued and as a result most of the sharps played him as well. Same with Thompson. Those two were close to 100% owned in some of the more competitive 10 man cash games I played. I’ve noticed that happening a lot lately, when my model is really pushing some of the lower priced guys, it’s not some weird thing that only I’ve picked up on, it’s something others know about as well. I’ll still try to put in some controls to avoid another Furykapocalypse but going forward, I’m not going to stress out so much if I’m going to be playing a mid priced guy in 7 or 8 of my lineups.
I have however, been putting a lot of thought into how to go about this. You see, oftentimes, the guys my model goes with on the cheaper end are usually marginally better than the next guy. If I’m measuring something on a scale of 1-100 and two guys at roughly the same price are 82.3 and 82.4 then the 82.4 guy is going to get selected each and every time even though there’s no real reason to believe he’s going to perform better. I’m thus going to add a miniscule randomizer to my model. if a lineup is produced that has some higher exposure players in it, I’ll run the randomizer moving some guys +/- 1% and spew out some more variations. If another one appears that has a slightly lowered exposed pivot then I’ll go with that one. A lot of me worries this is taking something pretty extreme and I definitely need to think more about the unintended consequences of doing this.
I’m currently in the midst of writing up a pretty lengthy blog about bankroll management. It’s one of the more commonly asked about things and it’s also one of the least understood. Much of what’s out there as common knowledge is stuff I personally disagree with. I’m in the midst of running Monte Carlo simulations to really beef it up so it won’t simply be my word vs group thought. I’m also considering deploying the simulator as a web app so people go there, enter in the type of games they’d wish to play and the simulator will tell them how frequently they’ll go broke and what amount played. Depending on how much time I have to do this, I may or may not get this out before the upcoming slate.
As always, please send me your feedback, negative or positive, it’ll be highly appreciated.