[mesadog]

Good day everyone.

I have been in the background of this forum for quite some time now reading and watching what some of you are doing in terms of data analysis etc. Let me start by saying some of the work being done here is very impressive (while some of it isn't.)

My partner and I are economics graduate student who have been working on model creation for NBA, MLB, and NFL. Recently we have applied the principles we have learned from the NBA to the WNBA. Our NBA model is very simple and concise however it gets the job done in a largely consistent manner. Over the previous season we went something along the lines of 58% and +120 units. This a good start for us in the NBA and we plan on improving the system this year. We're currently applying some of these changes to the WNBA and hopefully all turns out well.

Our MLB model, is needless to say much less simple than is our NBA model. For this we have done a great deal of model building utilizing various regressions from different angles etc. At the start we had great difficulty finding which variables were important, however over the past 2 months we have made great strides and have selected games at approximately 62% while gaining +60 units in that same time period. This includes several proposition bets that we have found are highly profitable when the correct types of analysis are applied. We are happy with this, however we feel this can still be improved... We feel as though we could get the model to where it is selecting closer to 65% or above.

The reason I am writing this is because we are interested in adding some outside help. Due to our rather busy schedules we don't have all the time in the world to be thinking about ways to improve / additional time for extra data collection beyond what we are already doing. I am curious if anyone would be interested in adding their input to the work we are currently doing. Ideally, we're looking for someone who can bring new ideas and angles to the table as to ways that we can improve the current model. If anyone would be interested please PM me and we could discuss what we are doing, and where we hope to improve/ move forward.

[TomG]

At the start we had great difficulty finding which variables were important, however over the past 2 months we have made great strides and have selected games at approximately 62% while gaining +60 units in that same time period. This includes several proposition bets that we have found are highly profitable when the correct types of analysis are applied. We are happy with this, however we feel this can still be improved... We feel as though we could get the model to where it is selecting closer to 65% or above.

You don't list the average break-even percentage of your plays. But I'll assume it's much less than ~60%. Right now you should NOT be trying to increase your win percentage. Instead, look to bet more games. This will lower your average edge.

Some friendly advice: You have a winning model. By enlisting the help of others you will likely give up much more value by sharing than you will obtain from others contributing.

[evo34]

Just curious: are your NBA bets all full-game bets, or do you bet halves? Also, what was your average unit size (that is, avg. "units" per game) to achieve the +120 units? Or more simply, what was your actual edge and what was your sample size?

[mesadog]

You don't list the average break-even percentage of your plays. But I'll assume it's much less than ~60%. Right now you should NOT be trying to increase your win percentage. Instead, look to bet more games. This will lower your average edge.

Some friendly advice: You have a winning model. By enlisting the help of others you will likely give up much more value by sharing than you will obtain from others contributing.

I'm not sure what you mean by this? Our winning percentage is ~60% in MLB. Volume is not the issue. We bet approximately 3-7 plays a day. Usually 1-3 Sides, and 2-4 prop bets. We currently are betting over/unders, team over/unders, and 1st inning score yes or no.

The following data is over the past 50 days since we started better record keeping...

60-38 Sides

70-39 1st Inning Yes/No

and the over/under's we have less than 25 plays due to the fact that we just discovered a solid method of attacking them, however those are ~62%.

As I said volume is not the issue. We are simply attempting to become more precise and move the sides up to an area closer to 65%. The prop bets are on the spot and I'm not a bit concerned with those. Here's the catch... if someone from the outside can help come up with other ways of attacking new projects etc... we feel as though we have the statistical knowledge to move our volume even higher. Our goal is to clear at a minimum +150 units in each of the 3 major sports. We're there in NBA, getting there in MLB, and we have a ways to go in NFL.

[mesadog]

"Just curious: are your NBA bets all full-game bets, or do you bet halves? Also, what was your average unit size (that is, avg. "units" per game) to achieve the +120 units? Or more simply, what was your actual edge and what was your sample size?"


We use a system where 1 unit = 1% of our starting bankroll. As we grow our bankroll (we are both in various levels of grad school so our starting roll was pathetic) we assume this will decrease.

Our sample size was... I don't know 65 games of the season (took the first 15 off to watch and learn then teams plus we needed data), and anywhere between 2-5 games per night with an average of about 3 I would say. We started out betting on only full games, and then slowly started to add in 1st Qs and 1st Half plays. So the volume increased by the end of the season.

[TomG]

Having a high W/L record is not necessarily a good thing and looking to go from a high W/L to even higher winning percentage will require passing up many +EV opportunities (or simply taking shorter odds).

If you can really hit at the rates indicated, your primary goal should not be to tweak anything with your model. Instead you should be aggressively trying to grow your bankroll. Understanding Kelly Staking and you'll be millionaires soon enough.

[CHUBNUT]

**** me, this scenario is played in every college across the planet. the "were smart college students and we can beat the odds" when are people going to realise that it takes hard work and intuition to beat the game, not mathematical formula.

yes you need to know your figures in a betting critera for maximum gain but after that its sharpness of mind that wins the day. I think the biggest problem with think tank thinking is it takes away the creative view our brains give us and instead of a view to an open mind, we get bogged down with figures.

[mesadog]

**** me, this scenario is played in every college across the planet. the "were smart college students and we can beat the odds" when are people going to realise that it takes hard work and intuition to beat the game, not mathematical formula.

yes you need to know your figures in a betting critera for maximum gain but after that its sharpness of mind that wins the day. I think the biggest problem with think tank thinking is it takes away the creative view our brains give us and instead of a view to an open mind, we get bogged down with figures.

Interestingly enough, this scenario plays out in any forum. There is always some guy who can't accept that others use tools that are beyond their scope, and try to neigh say because of it.

We are not some college kids... We are both 28+ and pursuing advanced degrees because that is where our lives have taken us. Please if you have nothing of positive nature to add to this conversation save it for someone else.

[mesadog]

I think the biggest problem with think tank thinking is it takes away the creative view our brains give us and instead of a view to an open mind, we get bogged down with figures.

I guess we should call Boston and tell them they can fire their sabermetric statisticians because according to CHUBNUT, it is useless and takes away from the creative process.

[evo34]

Not sure I follow. How were 120 units generated off of (~65*3=) ~200 bets? That is why I ask what your actual edge is (with units normalized to 1) -- i.e., what is your avg. pct. return per bet? (Bankroll has nothing to do with testing the quality of a model). Regardless of edge, unless you have some ridiculously robust (out of sample) backtest, I wouldn't get too excited about hitting 58% over 200 games in your first season of live testing. I know that wasn;t your original question, but I am just pointing out that evaluating the quality of a model hinges on calculating your realized edge (and/or your expected edge based on predicting open to close movement) on a large enough sample to know it's not luck.

"Just curious: are your NBA bets all full-game bets, or do you bet halves? Also, what was your average unit size (that is, avg. "units" per game) to achieve the +120 units? Or more simply, what was your actual edge and what was your sample size?" We use a system where 1 unit = 1% of our starting bankroll. As we grow our bankroll (we are both in various levels of grad school so our starting roll was pathetic) we assume this will decrease. Our sample size was... I don't know 65 games of the season (took the first 15 off to watch and learn then teams plus we needed data), and anywhere between 2-5 games per night with an average of about 3 I would say. We started out betting on only full games, and then slowly started to add in 1st Qs and 1st Half plays. So the volume increased by the end of the season.

[bztips]

evo is exactly right. Hitting 58% over a couple hundred NBA games or 62% over 2 months of a baseball season tells you very little about whether you have good models, or whether you've just been running lucky for awhile. You need to do some extensive out-of-sample testing.

Nevertheless, if you want a quick and dirty statistical test of whether your selections thus far are significantly different from random picking, take a look at the post by Ganchrow from a few years ago showing step-by-step how to compute a Z-score to evaluate your picks.

[evo34]

Here is that thread:

http://www.sportsbookreview.com/forum/players-ta...r-success.html

[mesadog]

I'm done posting in this thread. I don't need to convince the people of this thread on whether my model works or not. Fortunately that's not how I make my money. If you are interested in learning more about what my partner and I do, and are willing to do a bit of data collection to attempt to prove any theories you may have contact me via PM.

[mesadog]

evo is exactly right. Hitting 58% over a couple hundred NBA games or 62% over 2 months of a baseball season tells you very little about whether you have good models, or whether you've just been running lucky for awhile. You need to do some extensive out-of-sample testing.

Nevertheless, if you want a quick and dirty statistical test of whether your selections thus far are significantly different from random picking, take a look at the post by Ganchrow from a few years ago showing step-by-step how to compute a Z-score to evaluate your picks.

Lol had to make one last comment... as I said we are doing advanced level statistical analysis. I'm pretty sure a z-score calculation is within our grasp.

And you think we haven't done out of sample testing? How do you think we came up with our model? I don't need a bunch of people telling me how to derive a good model... That's not why I came and posted on this forum.

[bztips]

Gee, sorry I didn't pick up on your knowledge of "advanced level statistical analysis". Simply stating that you're an econ grad student doesn't NECESSARILY mean you're a stats whiz, but I'll take your word for it.

Frankly, you sound like an a-hole of the first order -- getting offended when posters who DO NOT KNOW your specific background make straightforward suggestions. A polite "thanks, but we've done all the backtesting", etc. would have been more appropriate.

Good luck on your modeling.

[ForgetWallStreet]

60-38 Sides

70-39 1st Inning Yes/No

Since you're clearly a genius who is above the good advice posted in this thread thus far (CHUBNUT excepted), I'm sure you understand just how worthless these figures are without the inclusion of average price bet.

[mesadog]

Look I'm sorry for losing my temper a bit... I understand you're trying to give good advice and what not.

I simply didn't come here looking for modeling advice. I just want to see if anyone is interested in getting involved in our project. My partner and I have a pretty busy schedule and could use some help with data collection.

Our MLB average price is I would have to take a guess... -125. That is irrelevant to what we do though (in a way)... In our modeling process we look for games that have the highest chance of winning and then alter the bet size depending on those figures. Really though, I don't care to bicker about if what I am doing is "significant" etc. I know what I am doing is good work, and if anyone else is interested please contact me. I would at the end of the day like to assemble a team of about 4-5 people, who are intelligent, and can each bring something to the table. With this, I believe we will be able to continue to improve and create massive profits (after all that's why we're all here isn't it?)

If you think about it... Vegas employs massive teams of statisticians with great minds. I think if we can pull together the best and the brightest from here, we can begin to do the same.

Now please, I understand some of you don't think what we are doing is "possible." And that is totally fine with me... as I said before, I'm not here to prove something to you. However, if you're intrigued by what we are doing and would like to get involved please contact me.

[ForgetWallStreet]

Look I'm sorry for losing my temper a bit... I understand you're trying to give good advice and what not.

I simply didn't come here looking for modeling advice. I just want to see if anyone is interested in getting involved in our project. My partner and I have a pretty busy schedule and could use some help with data collection.

Our MLB average price is I would have to take a guess... -125. That is irrelevant to what we do though (in a way)... In our modeling process we look for games that have the highest chance of winning and then alter the bet size depending on those figures. Really though, I don't care to bicker about if what I am doing is "significant" etc. I know what I am doing is good work, and if anyone else is interested please contact me. I would at the end of the day like to assemble a team of about 4-5 people, who are intelligent, and can each bring something to the table. With this, I believe we will be able to continue to improve and create massive profits (after all that's why we're all here isn't it?)

If you think about it... Vegas employs massive teams of statisticians with great minds. I think if we can pull together the best and the brightest from here, we can begin to do the same.

Now please, I understand some of you don't think what we are doing is "possible." And that is totally fine with me... as I said before, I'm not here to prove something to you. However, if you're intrigued by what we are doing and would like to get involved please contact me.

You may indeed have a great model, but it's painfully obvious from the statements in bold and others in this thread that you know little to nothing about how sports betting works. I'd fix that first before worrying about assembling a team and massive profits.

[mesadog]

You may indeed have a great model, but it's painfully obvious from the statements in bold and others in this thread that you know little to nothing about how sports betting works. I'd fix that first before worrying about assembling a team and massive profits.

Well I've been making steady profits for over a year now (which is just a start, and isn't going to stop). I hope to never know how sports betting works if that's the case! Good luck to you!

[mesadog]

Also are you implying vegas doesn't employ teams of statisticians?

[ForgetWallStreet]

Also are you implying vegas doesn't employ teams of statisticians?

Yes

[sharpcat]

The bright minds that exist here are probably already successful on their own or with a group of gamblers that they know have strong grasp on not only statistics and probability but also have a strong understanding of how the market operates. Only help you are going to attract here is from clueless posters hoping to piggy back off of you, it definately does not help when you start out by insulting a few of the brighter minds that you will find here.

[mesadog]

Yes

Then you sir are the one who knows nothing about sports betting. That is perhaps the most moronic and naive comment I've ever heard on this forum.

[ForgetWallStreet]

Then you sir are the one who knows nothing about sports betting. That is perhaps the most moronic and naive comment I've ever heard on this forum.

"Vegas" doesn't set the opening line so why would they need a team of statisticians?

[TomG]

If all you need is help with data collection, then you are probably in the wrong forum. Find a cheap data monkey / database administrator on eLance to help setup and maintain a databases for you. Then you can do all the real work modeling with that data.

[mesadog]

If all you need is help with data collection, then you are probably in the wrong forum. Find a cheap data monkey / database administrator on eLance to help setup and maintain a databases for you. Then you can do all the real work modeling with that data.

I'll check that out thank you.

[evo34]

It may not be why you came on the forum, but you did feel the need to tell everyone how "well" your model has performed. Which would be fine, except that your results (which you cannot even seem to articulate clearly) are not very conclusive. "NBA model is very simple and concise however it gets the job done in a largely consistent manner. Over the previous season we went something along the lines of 58% and +120 units." Really? Largely consistent? A 200-bet sample hardly establishes consistency, nor does it predict future performance well. I would ask how long you have been doing this, but you've already said less than a year. Five years from now, you'll likely have a different view about relying on a self-described "simple" model to beat an extremely liquid market (NBA) after observing a 200-game track record.

More than anything else, this quote:

"Our MLB average price is I would have to take a guess... -125. That is irrelevant to what we do though (in a way)... In our modeling process we look for games that have the highest chance of winning and then alter the bet size depending on those figures."

reveals that you genuinely do not understand the fundamentals of gambling. You're ignoring price, and adjusting bet size based on raw win pct. (not expected edge)?? You seem curiously fixated on improving raw win pct. without consideration of your actual edge. This is the surest sign of a beginner. Nothing wrong with being a beginner, but beginners should generally exercise a bit of caution and/or humility. E.g., the exact opposite of: "Well I've been making steady profits for over a year now (which is just a start, and isn't going to stop)."

Lol had to make one last comment... as I said we are doing advanced level statistical analysis. I'm pretty sure a z-score calculation is within our grasp. And you think we haven't done out of sample testing? How do you think we came up with our model? I don't need a bunch of people telling me how to derive a good model... That's not why I came and posted on this forum.

[Inspirited]

lol at sample sizes. Last year I started out 202-109(61%) or +20 flat bet units betting MLB for 1.75 months. Of course, I had no model. I was doing well and thought that some people made things too complicated through over analysis. Betting on baseball was easy! I was told that I would go broke on favorites at 2+2, but I really did not take much heed. Then came the dreaded 4th week of August. It took 9 days to go back to break-even after I went on a stretch of hitting only 40% of my games. Things completely reversed on me. I went broke on favorites and ended up only +4 units on dogs. (This year my faded picks are up 40 units; I'm so bad picking on my own.)

- Just my story of beginner's hubris. But you're in a better position with your modeling skills than I was OP.

[bettorjon]

this is what i do in soccer,

find a game with even odds for the under. wait for the over to raise even (usually 10-15mins) then place your bet.

always been good to me.

[Wrecktangle]

Mesa, the tank has essentially devolved in a pack of wild "numeric" dogs. They'd sooner rip you apart over some math nuance than engage in any sort of civil conversation, even about the math.

[mesadog]

Your concern is all much appreciated. Hopefully, I'll one day learn something about the nuances of sports betting.

[specialronnie29]

May I ask what are the main books you're playing at?

[mesadog]

BetJamaica.com

[BrianLaverty]

How are you betting yes/no 1st inning prop at bet jamaica when they dont offer them on 95% of games?

[Sportsfan800]

How are you betting yes/no 1st inning prop at bet jamaica when they dont offer them on 95% of games?

You're right they only have props for about three games a day. They put them up late as well (Late Afternoon) after all the value is gone. From what i've seen, they have a tendency to juice the "No Score" side a little heavy. A shrewd bettor could probrably take advantage of that by finding good bets on the "yes a score" side.
I don't see how anybody can make a long term profit betting on the "No Score" there.