Hockey Data and Sports Betting

[Bear of Bad News]

I doubt Vegas/books run mathematical models to set their lines; in fact I know for sure that they don't. Their lines are a reflection of around ~10ish of their top oddsmakers opinions averaged out (for smaller markets, maybe just 2-3). Historical data is, for the most part, meaningless.

Some of my friends from school now run predictive models for major bookkeepers. Where are you getting your information from?

 

[EVBetting Site]

I think your view of sports betting is a bit too romantic Cantor gaming uses almost solely computers to make their lines. They had a report on E60 about them.

http://www.youtube.com/watch?v=oYaZjLm72L8

Even though it mostly talked about live betting it's still true for their lines before games. They may have a few people who oversee the operations but lines are set by comptuers for the most part nowadays. Pinnacle uses the same lines as Cantor and they're doing pretty good except with the FBI maybe but that's another story

We're past the days when people could gather around and make better decisions. Models have been refined over the years and big data has become much more available and people have made good use of it whether it's for sports betting or make better business decisions.

For some sports, this is the case, mostly baseball, football, and basketball.

For hockey and soccer, which are such low scoring sports, a mathematical model cannot accurately capture what they need it to capture reliably.

You can further see that there is strong human input on the smaller markets, stuff like tier 3 Norwegian soccer and professional video game betting, where there are no statistics or very few that just simply capture wins and losses.

I agree that for baseball, football, and basketball the only way to win in the long run (in my opinion) is to have an accurate math model and that is how their lines are derived. Not so much for soccer, hockey, props, and the smaller markets.

 

[Bear of Bad News]

For hockey and soccer, which are such low scoring sports, a mathematical model cannot accurately capture what they need it to capture reliably.

Please elaborate.

As I mentioned above, I have friends who make a good living doing this.

Why are they being paid if it's not useful? I'm of the opinion that Vegas wouldn't just give away money.

 

[EVBetting Site]

Does he model hockey?

It's possible for sure but it would just take such an advanced model. I've been trying to create one myself for the last 3 years. With the three sports I mentioned, a math model can be relatively simple and accurate because of clearly defined offense and defensive possessions. With hockey and soccer it is much less clearly defined as to what is an offensive or defensive possession and even what is or is not an effective/impact offensive or defensive play

 

[Bear of Bad News]

Does he model hockey?

Both of my friends model hockey among other sports, yes.

I've also been contacted about future opportunities in this area.

 

[EVBetting Site]

What books does he work for?

Because I know for Pinnacle, they primarily open their soccer and hockey lines (among other markets, like E-sports in particular) based heavily on the opinion of their head oddsmaker for that particular market (they call em dealers) rather than a math model. They open at lower limits. Then they rely on the market itself to correct itself to the "right" number. Their profits come from high volume vig, which is a brilliant business model that is very successful in the long run.

Their true work comes after the opening line is already set. Money moves lines but they have the power to dictate how far the lines move.

 

[Bear of Bad News]

What books does he work for?

I can't (won't) reveal that. Suffice it to say that they are material players.

Ultimately, you're correct that the line will adjust automatically to the right levels. That doesn't mean that additional money cannot be gained by setting the starting line more appropriately.

 

[Mathletic]

For some sports, this is the case, mostly baseball, football, and basketball.

For hockey and soccer, which are such low scoring sports, a mathematical model cannot accurately capture what they need it to capture reliably.

You can further see that there is strong human input on the smaller markets, stuff like tier 3 Norwegian soccer and professional video game betting, where there are no statistics or very few that just simply capture wins and losses.

I agree that for baseball, football, and basketball the only way to win in the long run (in my opinion) is to have an accurate math model and that is how their lines are derived. Not so much for soccer, hockey, props, and the smaller markets.

:s there's plenty of goals scored/events in hockey to make a predictive model. One easy way to verify this is simply through the "pythagorean expectation". Using goals scored and against you can "predict" how many games a team should win. It's true that in soccer a simple pythagorean formula won't work but with a little tweaking you get the same results.

 

[EVBetting Site]

I can't (won't) reveal that. Suffice it to say that they are material players.

Ultimately, you're correct that the line will adjust automatically to the right levels. That doesn't mean that additional money cannot be gained by setting the starting line more appropriately.

For sure their goal should always be to set the most accurate opening line possible. But whether they do it with a model or with strict opinion (which if you think about it, is no doubt a model as well that only exists in that dealer's head) I thought was the debate.

In reality it has to be some combination of the two right? Opinion based on some fundamental principles.

I think eventually we will see models be used for hockey but I really don't think we are there yet. There is just too much that is not captured by the layman and advanced stats that exist today. And of the stuff that is captured even, a lot of it is not understood correctly/perfectly either.

Like I said, I've been working on my own NHL model for 3 years now. I feel like I've learned some interesting things from it but there are also such obvious, critical flaws to it that my head just "gets" but the model doesn't. And I can't tweak it to compensate for these things accurately. It can routinely spit out lines that are 20 cents off or something when my head spits out lines that coincide with the majority of openers.

If your friends can accurately model NHL hockey then thats very impressive (definitely possible, I feel I can eventually get there as well or I wouldn't have wasted so much time trying) but I feel like our brain can accurately process and sort out the available information better at this point (due to lack of enough good/comprehensive stats). The best approach then is probably based off a model but modify it based on your opinion, agree?

 

[Bear of Bad News]

The best approach then is probably based off a model but modify it based on your opinion, agree?

I've never said that opinions aren't important. But your suggestion earlier in the thread is that numbers aren't used (and remember the forum that you're having this discussion in).

Educated opinions are critically important.

 

[Mathletic]

For sure their goal should always be to set the most accurate opening line possible. But whether they do it with a model or with strict opinion (which if you think about it, is no doubt a model as well that only exists in that dealer's head) I thought was the debate.

In reality it has to be some combination of the two right? Opinion based on some fundamental principles.

I think eventually we will see models be used for hockey but I really don't think we are there yet. There is just too much that is not captured by the layman and advanced stats that exist today. And of the stuff that is captured even, a lot of it is not understood correctly/perfectly either.

Like I said, I've been working on my own NHL model for 3 years now. I feel like I've learned some interesting things from it but there are also such obvious, critical flaws to it that my head just "gets" but the model doesn't. And I can't tweak it to compensate for these things accurately. It can routinely spit out lines that are 20 cents off or something when my head spits out lines that coincide with the majority of openers.

If your friends can accurately model NHL hockey then thats very impressive (definitely possible, I feel I can eventually get there as well or I wouldn't have wasted so much time trying) but I feel like our brain can accurately process and sort out the available information better at this point (due to lack of enough good/comprehensive stats). The best approach then is probably based off a model but modify it based on your opinion, agree?

You have to remind yourself that it's not because we can't find something that this thing doesn't exist. You only have to follow http://www.hockey-reference.com/ they publish % each team has of winning the game. Granted they're not perfect, and tend to favor underdogs, it's still a good example that you can give good predictions through mathematical models.

 

[Bear of Bad News]

It can routinely spit out lines that are 20 cents off or something when my head spits out lines that coincide with the majority of openers.

To this point, if your algorithms' opening lines are 20 cents off from the consensus, in which cases does your algorithm produce results closer to the closing line than the opening line? (And in which cases does it produce results further away).

If your algorithm merely replicates consensus, then it has little marginal value (although it's interesting in and of itself). You can't build something better than consensus if you merely replicate it.

STATED DIFFERENTLY: if your model is different from the consensus, it that because of a flaw in your model, or because of something interesting?

 

[EVBetting Site]

To this point, if your algorithms' opening lines are 20 cents off from the consensus, in which cases does your algorithm produce results closer to the closing line than the opening line? (And in which cases does it produce results further away).

If your algorithm merely replicates consensus, then it has little marginal value (although it's interesting in and of itself). You can't build something better than consensus if you merely replicate it.

STATED DIFFERENTLY: if your model is different from the consensus, it that because of a flaw in your model, or because of something interesting?

Flaw in model

It's interesting for sure but very flawed. It doesn't even model consensus I don't think.

If there were three tiers with "true" being the highest and consensus 2nd, my model would be 3rd tier.

Getting to the 2nd tier would be very valueable and interesting in and of itself. I can't seem to reliably get there even. I haven't done work on it for a while now because I don't even know where to start or how to modify it so there's that.

I have a hard time valuing many things. It's not even close to being a finished product (I don't bet with this model of course, it's mainly just for personal interest). For example, how much is Gabriel Landeskog being out for tonight's game vs Edmonton worth? 2 cents? 5 cents? 10 cents? The market thinks its upwards of 10 cents apparently. My model would have a very hard time giving Edmonton 10 cents of value simply because Landeskog is out. Things like this really screw my model but my head definitely gets the value of a player like this and it doesn't seem too far-fetched to me that he is worth 10 cents in this kind of matchup on a team like Colorado that is lacking depth wingers.

 

[Mathletic]

Flaw in model

It's interesting for sure but very flawed. It doesn't even model consensus I don't think.

I disagree. SBR presents what side people bet on for each game http://www.sbrforum.com/betting-odds/

You'll see that there's plenty of games in which the market bet 60% on a side. Oddsmakers are confident that they're better able to predict the outcome of the game than the market. Of course you don't want to stray to far either because you can lose a lot of money that way i.e. the 2012 Super Bowl.

Maybe I midunderstood the pevious question though :s

 

[Bear of Bad News]

Stated differently, my initial question would be:

If your model projects an opening line of Vancouver -145, and the consensus opinion is Vancouver -165, how often (and in what circumstances) does the closing line:

End up closer to Vancouver -145
End up closer to Vancouver -165

 

[Mathletic]

Stated differently, my initial question would be:

If your model projects an opening line of Vancouver -145, and the consensus opinion is Vancouver -165, how often (and in what circumstances) does the closing line:

End up closer to Vancouver -145
End up closer to Vancouver -165

I don't have the answer to that question but my guess would be that the oddsmakers would move their line to -165 to balance money on each side to a certain extent. Maybe they wouldn't want more than 60% on one side so in that case their line would move until the money is evened out.

 

[Bear of Bad News]

I don't have the answer to that question but my guess would be that the oddsmakers would move their line to -165 to balance money on each side to a certain extent. Maybe they wouldn't want more than 60% on one side so in that case their line would move until the money is evened out.

That's not my point.

My point is that: if his model predicts Vancouver -145, and the consensus predicts Vancouver -165, if the line moves to Vancouver -145 by gametime, then the model is picking up something new.

Stated differently: it's not necessarily bad that his model does not replicate consensus.

 

[Mathletic]

That's not my point.

My point is that: if his model predicts Vancouver -145, and the consensus predicts Vancouver -165, if the line moves to Vancouver -145 by gametime, then the model is picking up something new.

Stated differently: it's not necessarily bad that his model does not replicate consensus.

oh ok, I agree

 

[EVBetting Site]

Mathletic, no offense but I don't think you are really following the conversation.

Taco, no my model doesn't nail closers either, it's just plain ineffective in my opinion. Closers are simply the opening line modified for lineup changes (injuries or otherwise, goalie change is the most common) between the time the opener is posted and the puck drops. I am 99% sure on this. There are maybe very rare cases where this is not so (this being a derivative of openers being so sharp to begin with no doubt). With how fluid the Internet has made the online betting market now, there is almost never going to be a case where literally 10+ books all post the same off line and it is hammered 10 cents by bettors.

The crux of the debate is that I feel that a math model is not as effective as a brain in betting NHL hockey. Now, I can accept that I may just not be able to accurately formulate what it is I want to formulate into an algorithm (and i know this is true too) but I really feel like there are not enough statistics out there right now to be able to do so. There are certainly things I want to add that are just not out there available right now.

I'm not really even sure what we are discussing or debating anymore to be honest. If you can 100% confirm that NHL lines are being set by a series of math models then I would be very interested and would like to share/pool ideas for a starting point on developing something similar

 

[RoyalOrange]

I think that you have taken the conversation to where we are talking about developing a model to compare and contrast with the lines that are being put out by oddsmakers. While that is a very interesting topic itself, I thought the spirit of the thread was to discuss anything that would give an edge to people that are betting on hockey, or perhaps just a different perspective on how to view hockey in a gambling light, especially in this unique season.

You do touch on the human aspect and the mathematical aspect, which I do think is pertinent to the topic of the thread. There is rarely something that is as easy as "keep betting the total points under" that will consistently make you money. However, there may be statistical trends that do occur in certain situations, but you must know the game of hockey and watch quite a bit of it to understand when to predict when these trends will occur.

Side note: I think that 10+ books put out the same line that ends up getting moved by bettors, on a pretty regular basis. Sometimes the public and the oddsmakers just think differently.

 

[EVBetting Site]

It moves, sure (every game moves pretty much), but 99% of the time there is a logical reason for the move based simply on lineup changes. What I don't see happening is 10+ books coming out with an off number, something that is obviously wrong and that the book has the right (rightly or wrongly, I think its unfair personally) to cancel all wagers made at this off number. You do see this in some of the smaller markets, and some of them go uncorrected; that's what I was referring to.

Getting from the opener to the closer is usually logically deducible (of course the degree of movement is very complex and something I do struggle with). I find the opener is also very complex and I don't understand it at all (in the sense that obviously you can ballpark it based on common sense, but to truly understand it is another beast all together).

I do feel like I understand closers reasonably well though, and I feel like the majority of movement on a line comes from lineup changes, whether that's different personnel in or simply a different allocation of ice time within a given personnel set. There are a ton of factors that people use when capping a game that I feel don't hold much, if any, merit (stuff like revenge factor, etc). I tend to disregard those and focus in on how lineup changes (roster change/ice time change) would shift a line solely, and it tends to follow line movement a very, very high percentage of the time (I would ballpark around 75+%). The times it doesn't I would probably attribute to me incorrectly interpreting how different ice times or roster changes would affect winning %. (Whether this is accurate or merely variance, who knows but it does make sense in my head and it's reasonably consistent over the last two years) Again, how far the line moves is much more of a mystery to me and some insight on that would definitely be priority #1 or close for me right now.

 

[RedeyeRocketeer]

Fwiw, scoring isn't up this year. It may appear that way, but the overall goals for is in that 2.6 to 2.7 range that's been fairly typical. Tampa is doing something special, but otherwise not much new.

To succeed in betting on hockey, just like in any other sport, the object is to beat the closing line. I beat the closing line in with Flyers-Caps game today by something like 25 points. I still lost, but the loss matters way less than the fact that I successfully beat the close. How often you beat the close is more important than wins and loses on a sample size like this short season.

Also for what it's worth, I'm finding tremendous value in live lines right now. Perhaps the most in the sport.

 

[ChadSC]

Fwiw, scoring isn't up this year. It may appear that way, but the overall goals for is in that 2.6 to 2.7 range that's been fairly typical. Tampa is doing something special, but otherwise not much new.

I wrote an article on this recently.

http://www.sportingcharts.com/articles/nhl/average-goals-scored-per-game-in-the-nhl.aspx

Scoring so far this season is 2.79 per game (per team), which is the highest since '08-'09. But not by any huge stretch - I think unless something significant changes in the game we are in for 2.65-2.75 for the foreseeable future.

The charts in the article update each day so you can check back to see how the GPG changes over the season.

 

[EVBetting Site]

Fwiw, scoring isn't up this year. It may appear that way, but the overall goals for is in that 2.6 to 2.7 range that's been fairly typical. Tampa is doing something special, but otherwise not much new.

To succeed in betting on hockey, just like in any other sport, the object is to beat the closing line. I beat the closing line in with Flyers-Caps game today by something like 25 points. I still lost, but the loss matters way less than the fact that I successfully beat the close. How often you beat the close is more important than wins and loses on a sample size like this short season.

Also for what it's worth, I'm finding tremendous value in live lines right now. Perhaps the most in the sport.

Obviously the goal is to beat the closer, but that's a redundant statement. It's like saying the object of playing a hockey game is to win the game. With the knowledge that you need to win the game, it's not like you are any better equipped to actually accomplish just that.

 

[RedeyeRocketeer]

Obviously the goal is to beat the closer, but that's a redundant statement. It's like saying the object of playing a hockey game is to win the game. With the knowledge that you need to win the game, it's not like you are any better equipped to actually accomplish just that.

I'm just saying that some people track wins and losses like it's the most important thing. I'd much rather track how often I beat the close.