TomG, have you seen this paper re: TOI?

https://archive.is/OsI36

Thank you for the motivation, I will begin posting euroleague plays again.

Thanks for sharing this. If I understand it correctly, though, the author is trying to estimate TOI (a missing variable in the data for minor league players) using goals data. This makes sense since they are working with limited data. It's the same general idea behind using Corsi as a proxy for team time of possession which was a stat not previously tracked at the NHL level. For our needs, though, we want to project a player's future TOI which is a different problem.

I think I just need to do a better job projecting baseline ES TOI and then adjusting for PP/PK as the great minds here have alluded.

Well that's embarrassing. I assumed it was projecting future TOI, should've read closer.

Long time lurker here - 2 thoughts I have.

1. From the outside looking in the Syndicate is a bit intimidating and it seems like if you don't have some sort of model then its probably best to post your plays elsewhere and lurk to glean information.

2. I currently make MMA plays (D Poirier+250 gogo!) that I don't record or share because I'm essentially betting these on my gut with no data to back it up other then line movements and historical performance. I want to create an MMA model and start posting lines/picks as well as data snippets to stir the conversation (probably better in the MMA forum then Syndicate TBH) but I don't even know where to begin. Its almost like we need a Beginners Crowsource Syndicate for all of the more basic questions to learn how to create a model before we advance to the actual Syndicate. I've never used R or Python - only excel/access. I'm not even sure which program would be better to use for MMA other than I've heard you all reference R a few times. Maybe you and Rsigs teach a class to everyone on the basics one night via discord or something if that would be enjoyable for you?

I'd love to post in the Syndicate forum more but I feel like I don't bring enough value to the table yet and I bet a lot of lurkers are in a similar situation to me and don't know how to start.

We all use models. Some just use a spreadsheet or program. Intuition is a model that isn't yet explicitly defined. That can be a problem and I do think people benefit from taking the time to make things explicit so they can avoid cognitive biases and harness/automate more advanced analysis techniques. A good model will beat intuition. But a good intuition is often better than a mediocre model.

I'm just trying to say don't be intimidated because worst case scenario you learn more about spreadsheets/programming. This thread is still a great place to get started.

Perhaps it is not time on ice, but other factors, such as a player's taste for board games such as monopoly, that determine their SOG performance. May I, humbly, suggest that you consider taking that into account? Love of a strategy game can provide good mental exercise, keeping a player sharp on the ice.

https://twitter.com/Avalanche/status...23511755608066

i took the plunge and drove to CO to test out the new DESK PRISM SPORTS live platform

initial reaction: it sucks

their MLs match other books, but their spreads and totals are way off of market and my predictions.

sounds great right? the odds are up for like 5 seconds and then are taken down and replaced with a lock.

i'd say during the LA/MIL game the amount of time you could actually live bet was 3-4 minutes of a 3 hour game.

any time i tried to place a bet the lock would come up before i could hit submit.

very quality product and makes me happy i drove 95 mph all the way across utah in a day to try it out

just staring at this screen the entire time lol

nice they just opened a total at 235.5 that pinnacle/BM have at 226.5

and +3 that pinny/BM have at pick 'em and +1 respectively.

A couple peculiar lines:

Most Rushing Yards in Conference Finals
- The favorite: Jamaal Williams +250

Who will have most Rushing Yards?
Leonard Fournette -120
Ronald Jones -120

BUT, Lenny is +375 to lead the week in rushing while RoJo is +450, so seems to be at least some value in taking Fournette to pace the Bucs in rushing.

Lol isn’t their whole thing “markets that stay up and bets don’t get rejected”?

Maybe that was the Circa interface being the problem? Or is DeckPrizm the front-end also?

Totals 9 pts off is lolz.

I'm guessing the way it works is

Someone at Circa opens/closes the lines. Uses the Deck Prism API to get the current lines.

Either they didn't have lines set in the API or the Circa person forgot to open them / was slow or some combo of both.

I'm interested to see what happens tonight with many games going off at once vs. last night as a slow night.

I just think the big issue is what population does it serve? Casuals will load up superbook or equivalent and jam in live props and bets. They're not gonna sit there waiting for a commercial and hoping the lines open up.

And pros are either going to pick off horribly bad lines or use BM/PN which have higher limits and are open for a majority of the game.

The only space is fulfills is being a worse version of BM/PN live but legal.

I remember betting during commercials on 5dimes back in 2001. It's not some novel concept. The BM/PN model is probably the best. Lines up most of the game. Lower limits when the game is live, higher limits during breaks.

The B365 model appeals to casuals with 900 derivative bets + props but they'll limit winners.

They're really neither while also not contributing anything new to be like hey i'll use that.

I think the real big issue is the people who designed it aren't sports bettors. They role play and act like they are, but they're a business trying to sell a product. While also framing it that "this is what people want". It's like did you do market research? I'm guessing if you did no one would want live lines that are only open a couple times during the game and are only the traditional spread/total/ML bets.

RIP my first NHL faceoffs bet MIN over SJ. At least the steam agreed with me, so I can eat CLV for dinner.

The sad thing though is I am watching the MIN/SJ game and Bonino (MIN) won the face off clear as day and somehow nhl.com came to a different conclusion.

Nice. Time to lawyer up.

I’ve been lurking since the lockdown started. Definitely learned a lot, thank you.

Could anyone suggest a heuristic to help with deciding whether a model has stopped working, before finding out the hard way by giving back profits? For example do you stop betting your model once its error metrics (mae, rmse etc.) start running worse than the market, or when you stop seeing CLV?

I recently tried taking on darts moneylines. At first, the market closing lines had a log loss of around 0.63, and my model was at around 0.61, with +ROI in the backtest results, so I started betting the PDC Home Tour III. Those matches ran about 1am to 8am, so I tried going to sleep when it gets dark here in Vegas and waking up at 2am, but the schedule was rough. I wanted to automate the wager execution, but I don’t have (legitimate) access to a sportsbook with an API, so I just stopped betting it after several days. When the Home Tour III ended I re-ran backtests on all the matches (using the closing line) and the wagering results ran positive. But the model log loss ended up 0.02 worse than the market. I assume the model was just lucky to show a +ROI then?

I was listening to the Tennis Abstract podcast, and the host was talking about doing research on matchups. He said matchups have very poor predictive value relative to other predictive methods (like elo rating) until there's like 40+ previous matchups between two players. He didn't give any more detail about it.

This got me thinking about how the markets almost always make adjustments for previous matchups because it's generally believed to have predictive value, especially if the last meeting was within 4-5 weeks and on the same surface, the line always favors the winner of the previous match more so than in the previous match.

Has anyone done research about the efficiency of lines based on matchup history and/or one recent previous matchup?

I got a bit of a refund on that because my bot spit out preds because Bonino played for them last year and I didn't update my sheet, so I bet them. Checked the logs and realized my error in time to lose money, but Nashville won anyway and it was a marginally good bet anyhow.

There are a few scorers that I think guess at it, I've noticed this as well. Sometimes you're a coin flip laying -115, it's the concession to the house.

Congratulations and cheers to all LAK bettors.

Did we ever get a reply from that guy sigs paid to ask about the faceoffs?

Sorry I can’t add much value yet but hopefully can in the future. Apologize if this is not the place to ask but any recommendations on how to figure out what stats etc are predictive and which are just noise. Any books or resources about this? Thanks

I didn’t know about “Tennis Abstract” so thank you for mentioning it.

When I read your post I thought of Mark Glickman, a contemporary chess playing mathematician who designed the Glicko and Glicko2 rating systems, which consider the recency of a match when rating a player, but his system also assumes one will use a personal computer to perform the more complex rating calculation. The Glicko system only updates skill ratings at a set period (such as every week or month; his suggestion is that the period be set to include 10-15 games), and if someone didn’t play during a rating period, the variance metric of their rating will be increased accordingly, to try and account for a period of “no practice” and thus more uncertainty about their skill.

When I was betting into the PDC Home Tour III darts moneylines I used an ensemble of models, which included ELO. Although ELO alone could not “beat the market”, its performance did improve by including a calibration period. For the first “n” matches for each player in my database, I used a high K value, and after that I lowered the K value, and this method seemed to help the ELO system converge faster onto each dartsplayer’s actual skill rating. You raised a very good point about “sample size” (40+ per player)… I’m going to try and expand my database and see if I can come up with some decent projections for the upcoming darts Masters on the 29th.

I don't recommend this book in general (I didn't get much out of it) but since you're talking about ELO and Glicko I think it should be mentioned.

Who's #1?: The Science of Rating and Ranking Paperback

It's a graduate-level look into different rating methods. Rare to find that kind of info all in one place. Elihu Feustel's review is spot on IMO.

Tyler Johnson didn’t respond to the request so it was auto cancelled. I’ll try to find someone else. Every other hockey player I saw on there was old or doesn’t do face offs.

This week went pretty good tho so maybe we don’t need him