The Rotoquant

Describing and predicting the world of sports in the language of mathematics

Month: April, 2014

Picks for the Wells Fargo Championship

Last week in the Zurich Open the machine had Seung-Yul Noh projected to post a top 15 finish. His odds to win were anywhere from 100-150 to 1, making a bet on him a seemingly lucrative proposition (his estimated odds to win from the short-history of the algorithm were more like 20-1). I had been fairing quite poorly over the prior 3-4 weeks – even with Kuchar accurately projected to win for the RBC Heritage. I didn’t actually place a bet on the Kuch because he was one of the favorites going in and therefore offered relatively unattractive odds. With a lack of data about the actual implied odds of the machine’s projected winners, I wasn’t sure if his win projection represented edge or was simply in-line with the oddsmakers. Since the machine hadn’t accurately projected the winner for the first 6 weeks I was tracking it, the odds of the projected winner actually wining were impossible to calculate without doing some questionable extrapolations from more frequently occurring observations (i.e. odds of being top 10 or 25).

Even now it’s not clear what having picked 1/7 winners actually means – 7 observations is still a terribly low sample size. It’s only clear to me that 1/7 is better than 0/7 winners, all things being equal. It also eliminates, for the short term, one test which would show that the algorithm was bunk; if it had gone 20-30+ tournaments without picking a single winner we could statistically surmise that the picks had almost no value. I imagine that we will need at least 30 tournaments, probably more, to compute with confidence what the implied odds are for these projections – but picking winners never hurts.

Throwing statistical significance out the window for a second, and making the brash assumption that the projections are a more accurate reflection of what’s likely to happen than the numbers put out by the oddsmakers – it wasn’t difficult to recognize that Noh represented value given his 13th projection and the 150-1 odds to win. If you force ranked the golfers by their odds to win last week, Noh would have ranked somewhere outside the top 50. There are plenty of tacit assumptions behind this method for defining value, and they should be explicitly recognized and quantified, something I hope to do over time.

For the short term it’s we can start with the most basic question, the only one I’ve looked into directly – how skewed is the game the odds-makers are offering us? Technically speaking all the odds to win for all the golfers ought to sum to exactly 1 + some amount of juice. To the extent that they sum to less than one the odds are skewed towards the better (not likely), to the extent they sum to more than one the odds are skewed towards the house (check). Based on the wide range of odds I see in various sports books and betting markets online, the ‘sum to 1 + juice’ principle means that, for golf in particular, the calculated juice is quite high. To see just how screwed golf betters are getting on average, I checked this figure for a few sites that offer odds.

For the european outfit betfair.com the odds sum to 1.172 (this is a realtime betting market, so these figures will change leading up to – and during – the event).  If you bet every single player in the field $100, you should expect, on average, to win (1/1.172)  * $100 or $85 in return. Already that makes golf betting more juice-prone than other sports betting markets, and the explanations for it are simple enough: golf betting has long odds, so ‘the house’ can potentially experience wild fluctuations if someone happens to place a large wager on a long-shot and win. Also, and more realistically to blame for the egregiousness of the odds offered, is that I doubt golf bettors have thought much about this notion… it’s more difficult to add up 150 different fractional than it is to look at a spread number in the NFL and recognize that, given this number of points, the outcome of the game should be a 50-50 coin flip, or look at a money line number of -105. Golf betters pick one possible winner out of 150, it requires more mental gymnastics to consider the implication that, out of that 150, someone has to win.

But this 1.172 figure is by means an indictment of Betfair. By comparison, the not-market-driven Bovada.com’s odds for the Wells Fargo Championship add to an astonishing 1.544… Now that is some seriously insurmountable odds-flogging.

At any rate – Bovada is the place I’ve chosen to test my picks… if I can turn an RIO there then anything should be possible! I ultimately placed a bet for Noh to come in the top 5 at 18-1 odds, not too shabby, but I’m still kicking myself for not having him to win (on Bovada he was probably more like 80-1, but still a handsome reward). This brings me to a broader point about golf-betting, particularly golf-betting to win; even when you think you have edge you will expect to lose in excess of 95% of your bets – and that’s if you’re picking mostly favorites.

I’ve historically focused all of my attention towards the analytics required to accurately predict sports outcomes. I’ve admittedly hitherto payed almost no attention to the ‘mechanics of betting’, and how someone best turns a bankroll + some accurate projections into an ROI. There’s a ton of excellent research out there about what percentage of one’s bankroll is appropriate to wager for an event with given implied and actual odds, for some amount of perceived edge, etc. It’s all very exciting stuff – and as I consume it I’ll be sure to regurgitate some of it here.

Keep in mind that to date, the machine has never been geared to project winners at all, instead focusing on projecting fantasy point production. In future iterations I might switch this focus. The two values (fantasy points and finishing tournament position) are certainly correlated, but fantasy golfers are disproportionally awarded for volatile play – an Eagle (-2 with respect to par for the hole) might be worth +8 fantasy points, while a Double-Bogey (+2 with respect to par for the hole) is only worth, say, -1 or -2 fantasy points. In terms of the players position on the leader board, however, an Eagle is just as helpful as a Double Bogey is damaging – so there is a potential non-trivial reduction in noise possible by just focusing on final tournament place instead of fantasy points.

At any rate, it’s been weeks since I’ve posted anything more than the figures here, and this little diatribe isn’t much of an improvement in terms of blog contents. Just wanted to point out that I’ve not totally checked out – I’m thinking of new ways to leverage/test the machine’s picks in the context of both DFS and sports betting more generally. That being said I am still quite pinched for time – and have opted for earlier production and publishing of the picks over the prior set of notorious comparisons. Perhaps a future iteration will have a betting-centric focus, comparing them to the oddsmakers… we shall see.

Good luck this week for the Wells Fargo Championship… on with the chlorophyll!

Projections for the Wells Fargo Championship:

Picks_Wells Fargo Championship_2014

 

Some notes on these picks…

Bill Haas is suffering from a wrist sprain he suffered two weeks ago, that caused him to withdraw from the RBC Heritage (he was one of my picks, blast!). The injury was not golf related – he suffered it on a fall down some stairs whilst handling a family emergency between rounds 1 and 2 of that event. That being said – his odds to win are exceptionally high for a golfer of his caliber, betfair.com listing him at 55-1… He played golf on Monday with the likes of Matt Kuchar and Harris English in his own charity event, and I haven’t heard reports as to whether his swing was impacted by the injury. I consider Haas a high-risk high-reward play in both DFS and traditional betting given his inflated odds and the likelihood that people will see a WD for his last outing and stay away. Anecdotally Noh had WD’d from the tournament immediately preceding his win last week – not all withdraws are created equal, and not all of them are a death sentence.

If I had to pick a ‘Noh’ for this coming week it’d be Pat Perez (80-1 on Betfair, 66-1 Bovada) or Cameron Tringale (130-1 on Betfair, 80-1 on Bovada… see what I’m dealing with here?!). Good luck to all – talk to you next week.

Picks for the Zurich Classic

Last week the machine had its first correct pick to win in Matt Kuchar. Gotta love the Kuch; independent of him being the top pick by the machine, I was happy to see him finally get the win-monkey off his back. Seemingly great dude, definitely great golfer. Now if I could convince myself to bet the guy the machine says is going to win, to win, I’d be doing alright. Alas, I got cute and had guys like Na who failed miserably, Haas who withdrew, and Tringale who was atrocious cluttering up my rosters… Lesson: you can be very wrong even if you’re right, the devil is very much in the details.

Again I’m pressed for time – and again you don’t care… here are the numbers, ya schmucks.

Projections for the Zurich Classic of New Orleans:

Picks_Zurich Classic Of New Orleans_2014

 

Good luck!

Picks for The RBC Heritage

Going again with the minimalist approach. If I don’t have time to write anything, it’s best not to try and fake it. I’ll let the picks do the talking…

Projections for RBC Heritage

Picks_RBC Heritage_2014

Good luck!

Picks for The Masters

Here we are – the biggest tournament of the golf year so far… and here I am – absolutely no time to work on this except for hitting the requisite buttons to perform the analysis. The numbers are sound, last week was solid, but I don’t have a second to spare to make this post any more elaborate than a stupid intro paragraph and the picks. If you want more in-depth analysis about the field, the story lines, the course, etc., use your Googler.

I figure you all just want the damn projections anyway (bastards!). So here they are…

Projections for The Masters

Picks_Masters_Blend2014

 

Notes: the cut line is at 60 + ties. Don’t trust Bernhard Langar – his projection is obviously a data issue. There are lots of data issues that can get you – so please do your own research to supplement these projections when making your rosters… and god speed ladies and gentlemen.

Hopefully next week we can have something more in-depth. I’ll say anecdotally that Notorious and the Machine both did quite well – it showed in all around low RMSE’s and an optimal blending alpha of .3, which is the highest it’s ever been, but the improvement achieved by that blending was minimal – meaning both sets of projections were near the minimum in their own right… Ok, that’s two more paragraphs (and an extra note) than I had time for. Good luck!

Shell Houston Open Picks, plus a look back at The Valero Texas Open

Again – time is short. Last week was a good week in terms of predictions, the Machine accurately predicted 2 of the top 5 fantasy golfers. Last week the axe swung the other direction in terms of externalities; Phil was on many of my rosters that were looking as if they would cash before his exit. Such is life. I took 15 minutes to check into the discrepancy between the RMSE and Cut% performance and I think I’ve identified a cause: the Machine often times treats new players,  champions tour players, or players with very little recent history too generously – they are sneaking inside the cut-line. User be warned – don’t just blindly set your rosters, these anomalies and others (i.e. injuries) must be manually taken into consideration. Ok, let’s get on with it…

Look back at The Valero Texas Open

The same caveats as always apply to this analysis.  

As usual the comparison picks are from Notorious at Rotogrinders.com, a very well respected name in the daily fantasy sports industry. If you don’t know what these graphs are showing (and you care) then just go to first fantasy golf post that’s hyperlinked above – it contains all the explanations of the relevant concepts used in this analysis. 

Below is a grid showing Notorious’ draft street points projections (DSP), the rescaled draft street points projections (RescaledDSP), the machine’s RescaleDSP and the actual RescaleDSP’s:

Picks_Valero Texas Open_2014_MFP_Grid

Looking at the RMSE, this week the Machine again dominated. This isn’t surprising at this point; the method is designed to make accurate predictions and it appears to be doing just that. The key tweak that must be made is to adjust for the low-data-quality cases of Web.com and/or Champions tour players, which with any luck will up the Cut% figures dramatically. Here’s the RMSE comparison:

Picks_Valero Texas Open_2014_RMSE_Grid

One of the errors I found in my short investigation was cutting out some part of the field unnecessarily. I’ve corrected for it and added it back, which makes crushing the RMSE more difficult (since there are more points to project). The optimal blending figure after this correction is roughly 10%, which is more commensurate with the history of this comparison:

Picks_Valero Texas Open_2014_Blending_Plot

 

The farther left you are on this line, the more of the machine’s projections you are using, the farther right here represents Notorious. So last week the *best* projections would have been a 90-10 blend of Notorious and the Machine.

Here’s the cut percentage for both of us:

Picks_Valero Texas Open_2014_Made_Cut_Grid

 

As well as the machine did at RMSE reduction, it was ousted by Notorious in the percentage of golfers correctly projected to make the cut. So the interpretation again here is that Notorious’ method did a better job of predicting the rank order in terms of fantasy performance, while the Machine did a better job at predicting the actual performances. The thing to note: Notorious doesn’t even bother projecting those ‘low-data-quality’ cases, each week the projection list of his method cuts off 10-15 extremely marginal or totally unknown golfers. Usually this turns out to be a wise decision, although there is high upside to finding a golfer with 500/1 odds who is, in fact, very likely to make the cut. Tough to do, but if you can do it repeatedly there’s plenty of money to be made.

Finally for those less mathematically inclined an idea of how many of the ‘top-n’ players were predicted correctly in each method.  Keep in mind all the caveats around who gets included in this analysis… If I get the comment ‘But, so-and-so didn’t finish in the top 10!’ I’m going to be very disappointed, dear reader.

 

Picks_Valero Texas Open_2014_TopN_Grid

I’d say we both had an pretty good predictive outing based on this chart – now that the data issue I mentioned is corrected it makes a lot more sense. To pick 2 of the top 5 fantasy golfers is excellent – especially considering one of my top 5 withdrew from the tournament (Phil… sometimes I hate you so much…Please don’t kill me). The winner-picking was a bust this week – but I did have outstanding 1$ wagers on Kuchar, Kevin Na, Jim Furyk and Pat Perez; picking 4 of the top 14 isn’t too bad and I have to consider last week encouraging despite the loss.

At this point everything is leading up the to Masters, where DraftStreet has already announced the first ever $100K guaranteed tournament for fantasy golf.  DraftKings answered the call with their $300K Spring Golf Classic, maintaining their spot in my mind as the premier golf offering in the DFS space.

There are indeed great things to come, but right now we’ve got ourselves an Arnold Palmer Invitational to dominate!

Picks for the Shell Houston Open

Here they are:

Picks_Shell Houston Open_2014

 

The red line here represents the cut, based off of these projections (the cut being returned to its normal 70 count). Keep in mind the caveats about injuries and player histories – perhaps Stricker is getting more credit than he’s worth from his performances in the back-half of last season? I’ll still be trusting the pick (he’s a timeless sage of a golfer).

Good luck everyone in your contests.  Talk to you next week.