Would love to get your thoughts on this reply during Rufus Peabody’s AMA on Twitter @matt_courchene and @will_courchene
What things would Rufus be incorporating into his model that are missing from DG?
I don’t know, we talk with Rufus from time to time and I think our models are fundamentally pretty similar. The fact that we don’t really do any manual tweaking during any given week hurts us. I’m sure there are useful tweaks that people make every week that we can’t incorporate given our process is pretty automated.
As for more substantial differences, maybe stuff to do with shot-level or hole-level data (e.g. hole-level fit). We don’t use that information, although I don’t think the hole-level actually matters (compared to course-level fit) except in some unusual cases. And the only shot-level information we use is information on hole outs. So there are certainly things you could look at that there. But overall I don’t think there are any gaping holes in what we do anymore. Despite what some people say (not Rufus, but others) there are no real secrets out there for modelling golf. Just incremental improvements that hopefully add up to something meaningful.
4 Likes
Well I think the weakness of any model is well known, in essence they depend on data and therefore you are basically predicting the past. When it comes to predictions you have to predict the future.
We’ve seen many players lose or gain their swings overnight and it results in a 2-4 SG difference per round. Morikawa goes from losing 5-10 shots on the green for the week to being just fine and you’re just throwing darts on which week this is. Bryson is losing 2 strokes from 100 yards and in compared to when he came out of the break and looked like a +3 SG player for 2 months. Momentum seems to be a big thing too.
It comes down to whether you want to be the best modeler or the best bettor. Over the long run the best bettors will make the best predictions because they lose money very fast if they don’t. By the way bettors use models too, but the best ones know when to trust in the model and when not to.
1 Like
Do you think the model may over-estimate the strength of PGA vs. Euro fields? I have heard that criticism from some sharp folks. In the same vein, would you be betting the Make/Miss cut props this week given it is a mixed field, and the numbers could be confounded by field strength mis-estimation? Maybe just take it easier on them?
No, I don’t think we over-estimate the field strength of PGA/Euro. That one is easy to check using several different methods. For example, if we look at the last 10 years of model predictions, if we are over-estimating field strength there should be a relationship between the fraction of a golfer’s rounds on the European Tour and their performance relative to our model’s expectation. And there isn’t. And if there was… we could just make an adjustment to fix that (which we actually did have to do with the Champions Tour, as for some reason our score adjustment method doesn’t work that well with the Senior tour). We also provided some intuition and a simple exercise to check that our field strength estimates make sense in this blog.
As said above, there are things that our model just doesn’t take into account, but I’m confident the model fundamentals (adjusting SG, incorporating SG categories, updating skill within tournament, etc) are better than anything else out there. Not because we are especially smart, I just think that we’ve done a lot of careful work that is statistically sound.
More generally, I’ve learned to not take criticism seriously from people who don’t provide any actual reasoning or data behind it. If the weight of their argument comes down to them being ‘sharp’, that doesn’t quite do it for me.
2 Likes
As for the Make/Miss, it probably does make sense to lay off a bit on them, simply because they all are very correlated.
Thanks Matt. I really do appreciate that color.
In my experience being “sharp” has far more to do with speed than handicapping or analytical skill.
1 Like
And it goes without saying the results of your model speak for themselves. Thanks for all ya’ll do.
Thanks. But to Daniel’s point, it is tricky to compare the quality of a model or a bettor’s ability to price things by looking at betting results, just because so much is dependent on the odds one gets.
Staking is also of critical importance. Not just which players you take at what number, but how much you place on it.
Good point, my rule of thumb for staking is:
Stake + winnings = 1% of bankroll
So if you have a $1000 bankroll and you wish to back Rahm to win the Open at +900, you wager $1 to win $9. $1 + $9 = $10 = 1% of 1000.
This sounds like a small amount but it adds up quick when you are making 100+ wagers a week.
I wouldn’t spend the time to calculate your edge and adjusting your bet size. That time is better spent locking in your wager before the lines move. You must win the race or you lose.
Agree to disagree here. Betting in proportion to your edge is extraordinarily important. A Kelly Criterion calculator isn’t that difficult to build (or find) and implement into a tracking sheet. I do think you’re best served to pick a fraction and stick with it though, rather than mess with that part of the equation. Then determine your edge thresholds to play
1 Like
In general I bet first and ask questions later, trying to beat the line move. I don’t always win the race even with this approach.
I do make adjustments on the big bets, usually the huge chalk wagers. Most weeks there are at most a handful of wagers that fit the bill so you can take your time and bet big on those.
You better be sure on your edges though. The big chalks is where I’m most confident of my calculations. However there are times when you see +170 on a 3-ball that Datagolf claims has a 15% edge and you’re wondering whether the book or Datagolf is right.
Overbetting small or nonexistent edges is the fastest way to go broke.
15% is hardly a small edge, especially on a +170 3-ball. Not a big enough edge for a Finish Position, but more than above the playable threshold for 3-ball. You never KNOW your edge, that’s why utilizing a (consistent) Kelly fraction is prudent. This isn’t Blackjack, there are too many unknowns.
@DG_Fan - would love to know more about this topic as it’s something I struggle with every week.
Is improper bet sizing going to negatively impact my ROI or is it just going to impact variance?
I’m using thresholds to determine if I should make a bet and I try to do some intuitive bet sizing but most of the bets I make are being placed at the max limit that the book allows.
I used to be good at math but I’m not anymore.
Continue to max bet or do I really need to get the sizing right based on a fraction of Kelly?
In my experience, Kelly Criterion is everything (Kelly in Vegas is nothing lol sorry)
2 Likes
Also, reading Ed Miller’s book “The Logic of Sportsbetting” is very helpful. Cap’t Jack has some good YouTube videos, some with links to resources (like a Kelly Calculator). I have no affiliation with either of these
You absolutely must use a fraction of Kelly. Not doing so will definitely negatively affect your ROI and variance.
1 Like
Use like 10% Kelly if you are not running the race
But identifying and wagering like 100-200 bets in 1 hour is not easy, LOL
Even at that speed I lose a lot of plays due to line movements