Early Games

WAS Evgeny Kuznetsov u2.5 -107
PHI Oskar Lindblom o2.5 +145
CGY Rasmus Andersson u2.5 -201
CGY Matthew Tkachuk u2.5 -110

70 CBB games today. Let's see some halftime bets and make RAS jealous.

Remainder of slate

TOR Tyson Barrie o2.5 +138
TOR Morgan Rielly o2.5 +108
DET Tyler Bertuzzi u2.5 -180
MTL Jeff Petry o2.5 +141
MTL Brendan Gallagher o3.5 -105
MTL Joel Armia o2.5 +118
FLA Aleksander Barkov u2.5 -123
STL Brayden Schenn u2.5 -167
DAL Joe Pavelski o2.5 +149
DAL Tyler Seguin o3.5 -106
VGK Shea Theodore o2.5 +144

17.99u risked.

Daily: 12-3-0, +10.84u
YTD: 106-114-0, -16.89u, 259.13u risked, -6.5% ROI

well done, a few more of those and these haters will be paying you for picks

i've been absolutely crushing nhl dfs this year, I've arguably vaulted myself from top 10 to #1 overall at the site, seriously considering dropping other sports and becoming an NHL specialist because I'm capped at about 2k of action per night on yahoo, been super nice because in NFL i'm barely profiting, only up about 1k on the year thus far

"As iron sharpens iron, so one person sharpens another."

Early run down.

DET Tyler Bertuzzi u2.5 -182

Pinnacle sometimes adds skaters throughout the day so there may be an afternoon update.

lol, yes, I expect it to be both frustrating and a pointless mess first round. Everyone just seems to copy Pinnacle within 30 seconds anyways, so I should probably just stick to chasing steam, but gonna give it a crack. Trying to dig myself out of a hole by the end of the month so my wife doesn't divorce me, but hoping by mid-dec to start posting lines.

How can you make a model that only does 2h? Seems weird no?

how is that weird?

Because its literally identical to how you model any other portion of the game.

All,

How do we go about regressing xG to actual goals? Lets assume we have some estimate of overall shot quality that we compute via an xG per attempt, we have a reasonably accurate estimation of path dependent variance (lol Markov Chain, gtfo). But xG isn't accurate since some people are actually better at shooting than others, but we're highly limited by contextual issues that may affect xG (team composition, etc).

For example Messi appears to be a pretty shitty passer based on xA, but is he really?

its pretty impressive that one could make so much money knowing nothing.


How can you make so good comments about the overall market and a shitty one like this?

So just don't regress them at all?

lol regressing xG.

I dont want to go in to much detail in a public forum, but is xG to goals a knowable variable? Is xG allways come to a mean in the end of a season, or some quantity of games?

Or you could join the guys who are past posting on 2nd half lines.

Are they bots with fast feeds or something and do they actually get away with it?

Everytime there is a big late line move on 2h lines, the game has already started and it’s in the direction of the team that scored first. .

I used to steam chase 2h but dont anymore. Too many big back and forth moves

Daily: 0-1-0, -1.82u
YTD: 106-115-0, -18.71u, 260.95u risked, -7.2% ROI

Monday plays

TBL Victor Hedman u2.5 -157
CBJ Gustav Nyquist o2.5 +138
VAN Brock Boeser u2.5 +123
VAN Bo Horvat u2.5 -105
CGY Rasmus Andersson u2.5 -157
MIN Matt Dumba o2.5 +125
STL Justin Faulk u2.5 -210
STL Brayden Schenn u2.5 -175
LAK Anze Kopitar u1.5 +146
SJS Logan Couture u2.5 -136
SJS Kevin Labanc u2.5 -193
SJS Erik Karlsson u2.5 -199

Pai,

That seems to be a really idiotic post. Many people regress xG or any other expected variable that is based on average stats based on player ability. Dunno how you thought my post is bad based on anything other than idiocy or a desire to gaslight which this thread is not about.

If you want to contribute please do, regression is irrelevant. Median is what is important. Maybe you are too stupid to understand my commentary but it is about knowable sample size when when we should regress and how.

I suggest building a data repository of 1H data such as injuries, field conditions, and weather. I use cbssports for injury data, but I don't think they go back prior to 2019. We could then build player VOA model for the available starters and go from there, either with simulations or non-linear aggregate projections such as a neural net.

Building a predictor of 2H spreads/totals based on historical data is pretty straightforward, which I would assume is a basis for openers, along with some subjective info that could be implied from the deltas.

Deeply, i will not extend the discussion. If you and many others use xG, fine. If they expect that variable to follow a median, fine.

I'm very disappointed that you would take this position of denigrating someone else in this thread built on collaboration without actually participating in any meaningful fashion.

xG is obviously an average shot that strips much of the basics of player skill out of the equation. DIPS/modern WA fielding metrics/wOBA/etc all are do the same thing to one degree or another. xFG% is an NBA stat that is literally the same thing.

Obviously xG is a cumulative stat, but its a great starting point for assessing relative shot quality and using that to compare performance relative to what would be expected which helps us understand player skill outside of their role. Or maybe you didn't understand what I was asking?

Regardless, I'm very disappointed you just came here to **** on someone else without actually contributing. If you wanna stomp your feet and say, determine their distribution of shots, the probability of them scoring on each one, the path dependence, then model it. k thanks great. I wasn't aware that is how you model literally any sport.

As you can imagine, its not in my interest to contribute much. Not on a public forum anyways.

What i can say is that xG is missleading, it only serves to somewhat explain the past and only that. It doenst strip anything of basic player skill. Of any player.

Saying that a team or a player has some score at xG that comparing that to the league median it should have 10 goals but only have 5, and expect in the future that to regress to the median is very wrong.

Sure, casualty it will give some good results, but its just that.

You get the bullseye in messi, why not extrapolate?


you can see similiarities betwen poisson and xG, inclusive both dont work.

this is coincidentally exactly how i feel about shots gained data in golf, admittedly not having read the book yet written by the guy who came up with the metric but it seems like just another way of counting results

Pai,

It seems you're pretty confused. Here is an explainer on the subject: https://www.americansocceranalysis.com/explanation

That clearly doesn't account for player skill. IE Why a header by Messi would be worth less than Ronaldo.

Beyond that it seems you clearly didn't understand my point. xG is a great tool for small sample sizes and to help explain the past, much like the metrics I gave above. However, some people over/under perform these due to a variety of issues that aren't encapsulated in the model. My question is simply how do we pull player skill apart from the other elements that are contextual with team usage. IE Where do we become fairly certain that a Ronaldo header from the same exact spot on the field is worth more expected goals than basically anyone else taking that shot.

I know what xG means but that metric have so many considerations upfront to get to that number, so much noise in it, that you lost all the litle signals that it could give you.

Its just like Poisson but with more math involved. Even for explaining the past.

This is the fun parte. Why would you wanna do that? So much trouble to get to an somewhat rough number with no pratical use ?

Umm... You pretty clearly didn't know what it meant 30min ago. So I'm gonna press F to doubt. It includes considerations like binning specific on goal locations? Including granular angle/distance data rejoined with shot type? I'm curious why you think this is poor historical data to utilize for the determination of whether someone is a better than "average" (since the data is self-weighting for better players due to the inherent nature of including all shots) at taking shots or creating assists.

WRT your second question, so you think that scorer ability is useless? Kinda weird way to model goals scored. I'm sure you have an elegant model that makes tons of sense.

Also, dunno why you harp on Poisson. Goal scoring isn't independent and never has been so pointing out the flaw in a Poisson model is clearly asinine. Is it good enough? Probably not. Can it be the basis for a winning model? Maybe? I dunno. I generally don't do the most ghetto thing.

I dont know why you think that i didnt know but whatever.
To continue the debate, i would have to disclosure ideas that i dont want to.

But to put it simple, you nailed it in the messi example, why not extrapolate that ideia ? Why not question the very basis of that stat?