How To Develop A Profitable Model

[VideoReview]

First off, as a first time poster to this new forum I think the creation of this new "Think Tank" discussion area is fantastic and kudos to whoever thought of it.

I have noticed, in general, that although there is tremendous help available here at SBR regarding the mathematics of betting, I have yet to see a clear discussion as to exactly how to develop a model with positive EV. I have come to the conclusion that it is simply a case of those with the information about creating positive EV models wanting to keep this specific information to themselves. To be clear, I am not suggesting that anyone should hand over a complex self adjusting algorithm with postive EV (but feel free to PM me if you have one ) but am mearly pointing out I have noticed that the math help just seems to diasappear when someone is getting close to the nitty gritty of it. In the words of at least one poster, they want those still searching for statisitically significant positive EV models to "do your own homework". This makes sense to me and if this is true, this means that developing a valid profitable model is much much harder (and probably more important to overall bankroll growth) than learning what amounts to bet (Kelly etc.) or how to find the best lines or other frequently discussed and reoccuring topics on the math of gambling that get precise responses, etc., etc.

So, just so I don't waste a great deal of everyones time with this thread when the answer was only an SBR advanced search away, is anyone aware of any explanations for my observations other than either:

a) Those with the knowledge of how to actually create a profitable model do not want to pass that critical link on.

OR

b) That no such knowledge exists on the SBR forum (the null hypothesis).

If not, I am perfectly fine with assuming option "a" is true and will continue to post with this assumption.

[Ganchrow]

Some information is just proprietary.

[Justin7]

VideoReply,

The knowledge is out there. I've personally spent hundreds of hours trying to develop models for various sports.

If you are serious about spending time on a model, I have two VERY strong suggestions.

1. Make SURE you have good injury info, and your model correctly accounts for that. Many "blind stat" guys ignore this.

2. Focus on a smaller sport. The more obscure it is, the worse the lines are. The worse the lines are, the more useful your model is. Don't bother with professional sports - when your computer model spits out a big line differential, your model is simply less accurate. High volume sports attract the most talented computer groups. Many people are making money at this, but it gets harder the more people you go up against. I'd model underwater basket weaving if there were a market for it, as long as no one else were doing it.

[VideoReview]

Some information is just proprietary.

Thanks Ganchrow. Based the candor of your answer and the distribution of precise math answers in this forum being significantly skewed towards yourself, I am pretty comfortable saying that option "a" is significant well below the 5% level. Just needed to clear the air for myself.

[VideoReview]

2. Focus on a smaller sport. The more obscure it is, the worse the lines are. The worse the lines are, the more useful your model is. Don't bother with professional sports

Good point. Although I have only taken a crack at making models for the big 4 professional North American sports, I have noticed I do better with NHL than NBA/MLB. Interestingly, I do find the NFL more profitable on a percentage basis but the number of plays I have vs. NHL is much much less. The only problem I can see would be liquidity as the maximum bet size for obscure sports tends to be much smaller so my positive EV would have to be much greater to justify the time and energy. I will keep it in mind though and may try something like WNBA or CFL as a step towards that direction to test the waters.

Thanks.

[Dark Horse]

If you want to learn about capping, I would suggest that you don't hang out here. Winners here aren't sharing a thing about winning models (except 'Baba' in the Spurs threads). This forum is about sports books, and there is no better place for info on books and free contests.

For some strange reason this place has attracted a disproportionate amount of losers. Not the right crowd to mix with if you're serious about winning. There are professionals here, but they tend to be very quiet. For your purpose you might find the statfox forum more useful for it regular sharing of angles and open experimental nature.

[Data]

I think that developing profitable model in general takes a few steps:
1) acquiring historical data
2) understanding of scores distributions (Poisson for soccer and hockey, Gaussian for basketball, etc)
3) understanding of the statistical research that has already been done (SABREmetrics for baseball, APBRmetrics for basketball)
4) using and refining methods and models from 3) with offensive and defensive efficiency being the cornerstones.
5) formalizing you proprietary angles and including them in the models from 4).
6) testing with data gathered at 1) and were not use in 5).

This is not a Holy Grail but something that would help me two years ago and saved me some time. Come on guys, let's share.

[acw]

Data,

Good to see that you are still active. Having any luck now?
I remember how you left the RX:

http://www.therxforum.com/showthread.php?t=139358

2) understanding of scores distributions (Poisson for soccer

I have noticed that some books indeed use poisson to calculate their correct scores payout, but I can tell you that poisson is simply (and documented) wrong for this!

[Data]

I have noticed that some books indeed use poisson to calculate their correct scores payout,

Once you combine all the scores probabilities you get the game lines as well.

but I can tell you that poisson is simply (and documented) wrong for this!

I am wondering where is that documented? No matter what I am very impressed with you that you can see simply that it is wrong, I cannot. For me, the logic behind it makes perfect sense, however I admit that my affair with soccer was very short and fruitless.

Still, what you saying is that while using Poisson is good enough for the books to make the lines it is not good enough for you. That is precisely my point: take a common model and improve it. If you can do this, you get a profitable model.

[Ganchrow]

I am wondering where is that documented? No matter what I am very impressed with you that you can see simply that it is wrong, I cannot. For me, the logic behind it makes perfect sense, however I admit that my affair with soccer was very short and fruitless.

The attached paper explores some of the issues with using Poisson for NHL scoring.

[katstale]

I don't come in this forum often, but Ganch you remind me a little of a little Russian guy called Izverg (that is a compliment). That aside, what the OP is talking abt requires some very talented nerd/geek software designer friends, a math person, a money man or two, and one well versed jock (this group at a minimum). The kind of info you are looking for is certainly not out in public forums. It also certainly exists.

If you post enough good stuff (out of the box type stuff)--someone will PM you and say "would you like to join another smaller--very exclusive club"!!

When/if this happens you are on your way. Good luck to you.

Hint: forget abt the injury report stuff, etc. etc---it is all abt the math.!!!

[curious]

I don't come in this forum often, but Ganch you remind me a little of a little Russian guy called Izverg (that is a compliment). That aside, what the OP is talking abt requires some very talented nerd/geek software designer friends, a math person, a money man or two, and one well versed jock (this group at a minimum). The kind of info you are looking for is certainly not out in public forums. It also certainly exists.

If you post enough good stuff (out of the box type stuff)--someone will PM you and say "would you like to join another smaller--very exclusive club"!!

When/if this happens you are on your way. Good luck to you.

Hint: forget abt the injury report stuff, etc. etc---it is all abt the math.!!!

Creating a predictive model requires determing which variables are important and assigning them the proper weighting. Some of these variables are objective. Some of them are subjective. Predictive power is not directly related to the complexity of the model or the number of variables. Personally, I think there is in inverse correlation to the complexity of the model and the number of variables with prediction accuracy. I have seen models that used a dizzying array of variables, and they were not very accurate. And I have seen models that use 3 variables and are accurate.

The best software programmers are not nerds/geeks but will have a background in either business or a hard science like physics.

[katstale]

Forgive my use of the term nerd/geek, but the people you describe would fall into my definition. Doesn't really matter their background--as long as they are "Picasso of programming" and can build a box to hold the thinking.

Second, you are correct that less variables are needed. What you read in handicapping forums will invariably take you down the wrong path.

[Justin7]

If you want to learn about capping, I would suggest that you don't hang out here. Winners here aren't sharing a thing about winning models (except 'Baba' in the Spurs threads). This forum is about sports books, and there is no better place for info on books and free contests.

For some strange reason this place has attracted a disproportionate amount of losers.

(snip)

Dark Horse,

I have heard from multiple sportsbook sources (when arguing about disputes) that SBR players as a whole are winners. There are a lot of sharp players here; yu just have to pique their interest.

[Dark Horse]

(snip)

Dark Horse,

I have heard from multiple sportsbook sources (when arguing about disputes) that SBR players as a whole are winners. There are a lot of sharp players here; yu just have to pique their interest.

Really? I'm willing to bet that you read very little of what is posted on the SBR forums. And that would be a wise choice.

[MrX]

I have heard from multiple sportsbook sources (when arguing about disputes) that SBR players as a whole are winners.

Maybe I'm missing something, but if this were the case, wouldn't they be crazy to advertise on SBR?

Paying money to attract +EV bettors to your sportsbook = bad business model.

[durito]

(snip)

Dark Horse,

I have heard from multiple sportsbook sources (when arguing about disputes) that SBR players as a whole are winners. There are a lot of sharp players here; yu just have to pique their interest.

Then how exactly is SBR in business?

[Justin7]

The attached paper explores some of the issues with using Poisson for NHL scoring.

Wow... that was hard-core and great!

[chemist]

(snip)

Dark Horse,

I have heard from multiple sportsbook sources (when arguing about disputes) that SBR players as a whole are winners.

In a dispute they would say that, wouldn't they.

[chemist]

Creating a predictive model requires determing which variables are important and assigning them the proper weighting. .

They sound like jobs for the AIC (or cross validation) and ordinary least squares regression, respectively.