Also for anyone building a PGA model I would really read and try to understand the Connolly paper. I can answer any math/stats related questions. They basically solved golf and published a paper in a journal to get 38 citations over 12 years instead of making millions.

This paper (especially section 4.1) can help you develop lines to win tournaments

http://www.columbia.edu/~mnb2/broadi..._rendleman.pdf

How can you calculate h2h matchups? Use your baseline skills from the first paper + ideas form the paper above + course fit + add a random component (sampling from the normal distribution is fine, i use my own based on some historical data i have but the results only differ slightly from the normal dist)

so each golfer has Baseline skills + course fit adjustment + random component

why the random component? golf has randomness to it. if you don't do it then you'll just get one outcome of the tournament. the random will allow you to generate a distribution of outcomes for all the golfers

generate a score for each round for each golfer. store their final scores

repeat x10,000 times (or however many you want. i do each round 10000 times because 10000 seems like a big number)

use the distribution function based on all your simulations to calculate probability that golfer 1 > golfer 2

or see how many times they win, in top 5, top 10, top 20, etc.

the idea behind this you can use for any head to head match-ups for really any sports. the only difference is how do you adjust based on event and how do you calculate the baseline skill.

i use it for mlb, nfl, nba props. nascar (tho that is a disaster), i did tennis once but the limits got cut so quick i gave up.

also one note about above.

if you're calculating top 5, 10, 20 to bet you need to consider whether or not the sportsbook has dead heat rules.

so i calculate top 5, 10, 20, etc. which i use for DFS and then a separate top 5, 10, 20 for betting those markets.

and the reason i don't use the normal dist for randomness is i think some golfers have higher variance than others (e.g., spieth, vaughn taylor). if i have enough data from one golfer I use randomness based on their historical data. if i don't, i use the normal distribution.

Sorry for the 900th post about how to predict, but you might be thinking - this seems simple why would there be any edge?

In each of those components you need to make choices and those choices are what separates your model from datagolfs from BM linesmaker in trainer rufus from pinnacle etc.

When calculating baseline skills the paper using a linear regression approach. Maybe you want to add a new fancy machine learning approach. The benefits of these new approaches is they can find relationships in highly non linear types of data. FWIW I think regression is fine for golf, other sports not so much.

Course fit. Same above. Some people don't believe in course history for a golfer. Do you include or exclude that? Morning/Afternoon waves if you're focusing on Round matchups. Most sportsbooks keep people in the same group together to avoid that but look at the lines BM posts. A lot of them (pre-covid) were actually golfers who weren't paired together and started at completely different times. Do you only include the same course or do you try to develop a similarity index across courses? And your similarity index, is it 1:1? Do you map the current course to all courses (even if it has a low or negative course fit) or do you only use the courses this weeks one is very similar to?

Is there correlation between the randomness you add to each golfers score or are they independent? Is the "hot hand" theory true?

These are all choices you make which influence your final score. The idea is we want to make better choices than the books there. How do you do that? Use statistics. Constantly re-analyze these things. Make adjustments based on long term data not short term.

10,000 seems low

probably why we got crushed last week or it could be valuing course history too much when the field was stronger than usual due to covid

i'll add another 0 for next week

Is this the Chalkiest lineup ever?

Koepka, Spieth, Bubba, Chez, Casey, Nesmith

For the random component, do some players inherently have more volatility than others? Would think someone who makes more birdies + bogeys would be more volatile than someone with the same average score who makes more pars. Could go deeper too as some skills (putting) are more volatile than others. I assume everyone has a pretty similar variance which is probably a mistake.

I've been incorporating birdie and bogey rate into my simulations. At first I tried using each players Par 3 birdie/bogey rates and same with Par 4s/5s. I feel that includes putting, which I don't want to do, since it's the least sticky stat out there. Guys get hot/cold with the putter very quickly.

Now, I'm using it to make a slight adjustment to birdie probabilities on par 3s, 4s, and 5s.

I'm also incorporating SG-OTT on par 4s/5s, SG-APP on all holes, and SG-ARG on par 5s.

I'm also throwing in a few extra stats to adjust those probabilities like Driving Accuracy (this week) plus Proximity to hole on expected approach distances. This week, guys will have a ton of wedges and short irons in to greens (course measures 6800 yards).

My simulations are very stat based, so good players who have played badly (Spieth/Koepka) are taking a hit. I'm still entering lineups and making final decisions, though. So, I can overrule the simulations if I choose.

It does help me find some cheap plays. Had Higgs and Ortiz last week. I did also have Reed and Matsuyama, though. I hit on Webb and Hatton.

yep tons. That's why someone like Vaughn Taylor (he's an extreme example and always pops up every week when I look why I use him a lot) is hard to figure out. He is top 3 (might even be leading now) in birdies in the last year but he's also something like top 10 in bogeys. And it's not like he goes one round all birdies then all bogeys, it just happens every round.

He's one of my highest variance players so when I generate the randomness for him his projections have a wide band.

BM Full tournament bets. They have the Scheffler/Conner bet we got earlier at better odds, so I re-bet it. That's going to be our big loser this week. I can feel it.

Tournament Matchup - Travelers - Paul Casey +126 - (Paul Casey vrs Abraham Ancer) RISK: 0.7u
Tournament Matchup - Travelers - Patrick Cantlay +118 - (Patrick Cantlay vrs Xander Schauffele) RISK: 1.49u
Tournament Matchup - Travelers - Webb Simpson -141 - (Webb Simpson vrs Brooks Koepka) RISK: 1.39u
Tournament Matchup - Travelers - Patrick Cantlay +105 - (Patrick Cantlay vrs Justin Rose) RISK: 2u
Tournament Matchup - Travelers - Xander Schauffele -126 - (Xander Schauffele vrs Justin Rose) RISK: 1.5u
Tournament Matchup - Travelers - Patrick Reed -132 - (Patrick Reed vrs Sergio Garcia) RISK: 0.58u
Tournament Matchup - Travelers - Marc Leishman +103 - (Tony Finau vrs Marc Leishman) RISK: 0.66u
Tournament Matchup - Travelers - Billy Horschel -144 - (Billy Horschel vrs Ryan Palmer) RISK: 1.27u
Tournament Matchup - Travelers - Scottie Scheffler +100 - (Corey Conners vrs Scottie Scheffler) RISK: 2u
Tournament Matchup - Travelers - Chez Reavie -104 - (Chez Reavie vrs Keegan Bradley) RISK: 0.83u

we may be on the wrong side of this scheffler/conners matchup

bet it at +105, went to +100, now it's +109

what do strokes gained pundits think

Makes a lot of sense. I'll need to do some more work rather than assuming everyone has similar variance in my model. Fantasy seems to reward the high variance activities so I'm guessing the guys who rank higher in DK pts than expecting scoring average are going to be the same ones that are higher variance, just need to quantify.

Off of pure strokes I have fair at Scheffler -130. Maybe a question of how to weight recent form since Conners has been ~20th last couple weeks and Scheffler was 55th & DNP?

couple fade the steam plays

Tournament Matchup - Travelers - Jason Kokrak -155 - (Branden Grace vrs Jason Kokrak) RISK: 0.59u
Tournament Matchup - Travelers - Jon Rahm -159 - (Jon Rahm vrs Brooks Koepka) RISK: 3.18u
Tournament Matchup - Travelers - Kevin Kisner +116 - (Kevin Kisner vrs Joel Dahmen) RISK: 0.7u


you can calculate the variance of the player by looking at the residuals of your model for every event. then when you sample the randomness instead of sampling from a normal distribution with mean 0 and variance equal to 1, use mean 0 variance of whatever the golfers variance is

Sigs do you do any 1st rd leader betting? Wondering if you know what % first rd winners end up winning the tournament.

Thinking of hammering

Mcllory +2500
Rahm +3300

Scheffler did WD pre-tournament last week at RBC so that could be an influence?

I think recency is in some people models. Lot of the good golfers from the last two events are getting steamed.

Couple more props. All but Mickelson are close, so only bet if you can get that or better.

Brooks Koepka MISS cut +240 (+212)
Bubba Watson MISS cut +185 (+141)
Rahm MAKE cut -400 (-501)
Justin Rose MISS cut +240 (+217)
Phil Mickelson MISS cut -125 (i have this at -180, the most off one)

I added more inputs (par 5 eagle rate and more proximity to hole) and reran some simulations. Incorporating critiakx2's question about variance, I'm looking at 400 simulated rounds (that's 100 simulated tournaments). I did 800 the first time around (that's 200 simulated tournaments).

I've entered lineups based on guys avg scores, minimum scores, biggest range of outcomes, and now lowest score if everyone plays 1 SD better than their average.

Cheap DK guys that showed up (I wouldn't go overboard with these guys):
Beau Hossler (6600)
Russell Henley (6800)
Luke List (6900)...I didn't incorporate any KF Tour data
Max Homa (7000)
Bronson Burgoon (6100)
Cameron Percy (6200)
Grayson Murray (6300)
Emiliano Grillo (6300)...I was on him last year when he missed the cut (I think)

Edit: I think I like Ryan Moore better than Luke List at 6900.

Ryan Moore has good course fit here, but he's also Ryan Moore. For course history I have him #1, then Brian Harman, Bubba Watson, Hunter Mahan, Charles Hoffman, Chris Stroud

Bubba is the odd one. He has a good history but his style is not a good course fit. He's low on driving distance and accuracy.


One more fade the steam:

Tournament Matchup - Travelers - Justin Thomas +100 - (Bryson Dechambeau vrs Justin Thomas) RISK: 0.85u

Brooks Koepka will also withdraw from Travelers after his caddie tests positive for coronavirus

https://golfweek.usatoday.com/2020/0...r-coronavirus/

PGA Commissioner is addressing the media in 15 minutes. Is golf over?

Consensus on golf twitter is that they're still going to play. Time for some contact tracing steam.

I personally think books should add new rules

If one player withdraws due to COVID the match-up is cancelled.

Apparently Bill Kraken, the guy from that Showtime reality TV Show about Vegas Davis, is on Rose over Cantlay. RIP that bet right

Koepka's caddy was negative on a re-test.

Why isn't the standard procedure to assume false positive and re-test before releasing results.

Sigs can you give a bayes theorem class to these morons.

I'd feel better barreling this **** if they did that. In theory it evens out but who wants that added variance and losing that way is going to rub n00bs the wrong way, which is bad for the game.

Can Koepka un withdraw now?

They need to start posting lines for this LPGA Japan event. Japan figured covid out somehow