Also, if a bet is 0.30 EV or higher like some of these T20 bets, is it okay to break the Kelly formula rule? Seems like we should be taking advantage of these big EV spots more even if the odds are like +1200 or more.
Yes you should bet only $42. The longshots will slowly drain your bankroll in between the hits so you want to bet small
As for the high EV plays, it depends. If it’s Kisner at +1200 it’s an obvious mistake so you might slam $100 on the play. But if it’s Hickok at +1200 then well it might be model error and not book error. In my experience the high EV plays don’t perform that much better than the lower EV plays.
This is where your personal judgment comes in. If the model EV is high and you feel it’s justified then you might make a 1.5X or 2X wager.
While the Top 5/10/20 markets have been good to me I have been getting destroyed on Matchups. Usually don’t bet unless i have 8 cents of value or more.
1 Like
My worst weekend of matchup betting as well. I would be interested to know whether there were any changes to the model for calendar year 2022 (or since the last blog update). I did quite well betting primarily matchups starting in the back half of last year, generating an ROI of around 8% on 1,025 bets at harmean odds of 2.37. Using exactly the same criteria, I am negative YTD (following the Byron Nelson) after 969 bets. Adjusting for the slightly longer harmean odds and slightly smaller sample size in 2022, there is only a 1.4% chance of that happening due to variance alone. Still strongly believe in the model, just trying to figure out whether this is something anyone else has been dealing with/whether I need to adjust my thresholds… harder to compare now that the betting results page isn’t being updated.
Where are you betting?
I think matchups are probably the hardest to win against with our model (as a few people have stated in this forum) for a couple reasons. First the market just seems sharper, as even if you don’t have a full-blown simulation model you can still price h2h matchups okay. And second, matchup markets seem to copy us pretty heavily. Just take this week… other than a few guys the market really thinks we are off on (e.g. DJ) there isn’t much value to be had anywhere. This copying problem is worse for matchups likely because we post our stuff on Monday even though books generally don’t post their matchups until Tuesday morning (B365, BetOnline being the 2 exceptions I know of who post on Monday). DK is an exception – since leaving Kambi they don’t copy us anymore it seems.
Regarding the model, there haven’t been any meaningful changes, just cleaning up a few things. Those changes are detailed in a couple of the posts in the model talk series.
I’ve quickly grabbed our results at various EV thresholds using all bets (matchups & 3-balls) from 2021 and 2022 separately. 2021 was a better year, no doubt. (“pb” is Pinnacle by the way… don’t ask why.)
total profits across all books in '21:
profits by book for optimal threshold (6%):
total profits for '22:
profits by book @ 8% threshold:
We will eventually write another blog that explores this more fully. I’m definitely more confident in the PGA betting than Euro, I think the SG data is sometimes inaccurate on the Euro tour and our course fit stuff is worse probably there as well.
One thing to keep in mind regarding your p-value calculations is that it’s likely assuming independence of bets. In reality lots of your bets are correlated in a few ways: betting against / on the same guy in the same event is an obvious one, but also betting against / on the same guy over a series of events is another more subtle correlation (e.g. we are always betting against Marc Leishman, so if the model is just off on Leishman the EV of those bets will all be over-estimated).
Anyways, hope this helps and I hope the betting results pick up. If you want to see any more figures from 2022 I can see what can I do. I’m just coming back from a bad flu right now (covid, probably), so my productivity is quite low.
5 Likes
Great post, matt, and get well soon!
1 Like
https://feeds.datagolf.com/betting-tools/matchups-all-pairings?tour=pga&odds_format=decimal&file_format=json&key=API_TOKEN can i ask what is wrong with this
You need to replace API_TOKEN with your key, which will be shown at the top of the API documentation page if you are a Scratch Plus subscriber. If there is an error with your url, the error messages you see by clicking the url are informative in most cases.
1 Like
Thanks a lot, really appreciate the additional detail. I’m scattered across the US books and the offshore ones - I travel a decent amount so it’s mostly based on what I can access on a given weekend. It’s certainly possible I’ve just had bad timing in terms of the offerings for wherever I played - statistically it’s likely that someone on this forum has, and it could well be me.
Your point about the hidden correlations is well taken too. We all know the risks involved with betting and like I said I still strongly believe in the model. Seems like for now it makes sense to stay the course; just wanted to make sure I wasn’t missing something. Feel better soon!
#1 - just as an FYI, a few of the values scraped on the miss cut page for BetOnline are incorrect… for example, it’s switched Matthew Wolff’s make/miss cut odds for some reason.
#2 - Matt’s comment about hidden correlations/Leishman led me to do a bit of digging on which players have had an outsized impact on my results YTD. In case anyone is curious, I figured I’d post them here. The numbers below assume constant bet sizing (I also calculated them for KC sizing - if anyone is interested let me know). Take these worth a grain of salt, as they are small samples and don’t include every matchup offered (just the ones on books I could access that week)… just thought it was interesting which players that the model and market have disagreed on frequently and how those have played out. Curious whether others have seen a similar pattern in their results.
Most Profitable to Bet On YTD (Name and Z-Score)
1.) Cameron Young 3.10
2.) Scottie Scheffler 2.82
3.) Matt Fitzpatrick 1.92
Least Profitable to Bet On YTD
1.) Cameron Smith -4.31
2.) Joel Dahmen -2.33
3.) Bubba Watson -2.27
Most Profitable to Bet Against YTD
1.) Paul Casey 2.21
2.) Webb Simpson 1.88
3.) Corey Conners 1.84
Least Profitable to Bet Against YTD
1.) Jordan Spieth -6.07
2.) Marc Leishman -3.59
3.) Collin Morikawa -2.98
Overall (Both Sides Combined) Most Profitable
1.) Cameron Young 3.32
2.) Paul Casey 2.95
3.) Dustin Johnson 2.84
Overall (Both Sides Combined) Least Profitable
1.) Jordan Spieth -5.83
2.) Marc Leishman -3.56
3.) Collin Morikawa -3.42
4 Likes
Cam Young is such a beast. Rest in peace to the Cameron Young trade.
1 Like
Gosh that’s rough with Spieth and Leish. Both are DG killers, always have been.
Spieth is Jekyll and Hyde, he’s missed 3 cuts too so when he’s off he’s really off
I think he might actually be more inconsistent than Bryson
Might as well play him to miss the cut and finish in the Top 5 every week. 50% chance your ticket cashes
Kevin Kisner is another Jekyll and Hyde player Datagolf seems to have trouble with
It’s either a Top 10 or a MC for him
Must be an absolute killer to have a position on Leishman where DG actually showed value on him then. Snatched 2.14x vs Simpson. Can’t lose, right? 
1 Like
“Least profitable to bet against” must be Davis Riley as of late. Always DG value against him even though his DG rankings is quite alot better than his OWGR. Not much to do when he’s killing it week in and week out. And no recession in sight.
1 Like
sigh… yep. Amazing how betting against a player enough can make you hate anyone.
1 Like
Jordan Spieth and Colin Morikawa are infuriating exactly for that reason
Whenever they’re in contention they’re like -40% EV
It’s not just Data Golf. I make my own adjusted odds using initial book odds and their in-running prices are still awful beyond belief
1 Like