Is 23,331 game enough of a sample size

[statnerds]

To begin to draw conclusions?

How would you safe guard against your data being rendered irrelevant by a constantly evolving market?

Or would you only focus on criteria you felt would be unaffected by the evolution?

No one ever made money following the crowd. Been trying to target areas that produce results with limited variance through market corrections.

Just a thought.

[Peregrine Stoop]

if you were a statnerd, You'd know how to answer this question.

Depends on what you are testing as to whether you have enough of a sample yet

[Duff85]

I'm sure someone better at the math side will hammer me for this, but its my opinion that seldom if ever do we reach the "long term". As you've alluded to, market adjustments and the constant evolution of both our own betting style and the market itself will render your older data irrelevant.

Also if you have enough data to conclude that you have been on the right track in the past - its exactly that- you were at that point in time doing enough to beat the market. It obviously doesn't guarantee that what your doing in the future will continue to beat the market. Thus I think its almost irrelevant to gather the sort of data required to figure out if you have been winning against the market.

I think in a lot of ways we just have to have faith in what were doing.

[Nookx]

statnerds you have to tell us what info you are gonna use from this 23,331 game sample to tell you if that sample size is statistically relevant and to what degree.

[the_mathman]

of course yes, it is.

tell us about it. what are you searching?

[statnerds]

Also if you have enough data to conclude that you have been on the right track in the past - its exactly that- you were at that point in time doing enough to beat the market. It obviously doesn't guarantee that what your doing in the future will continue to beat the market.

Like that a lot Duff.

It is the constant dilemma I face. You have to have some data to go, but how much is too much?

Like I said, I analyze some things to determine if any constants survive market evolution. Also depends on the size of the market, which goes without saying, but I did anyway just to keep anyone else from pointing out the obvious.

Another thing I look for is the ability to apply anything I've discovered, or think I have, to a suitable future sample. That sounds stupid. Having trouble articulating that if I have something I think was profitable over 23,331 prior events, I would want to be able to sustain any Variance through proper money management. I wouldn't expect it to be profitable over the first 200, 300 or 500 games. It might, but also might not. So I look for ways to manage my BR so that i can realize those potential profits over a few thousands plays. The problem with keeping it to ~1% of my BR is profits are lower, but I also avoid Gamblers Ruin to an extent. If I do go bust, it would mean whatever I thought was a viable angle just lost 100 Units. So I try to analyze the peaks and valleys and prepare for the worst.

First thing I thought of when reading your post was how NFL Home dogs used to be money. Not so much after a while. But then I thought maybe that was less a Market Correction than it was simply Variance over a season or two or three or whatever it was.

[the_mathman]

Hi Statnerds
the yours is a big dilemma.
Is good if you have a profitable method on a long period, but the problem is that we don't have an infinite bank.
So, the best money management, if you want resist on the long period is (unfortunately) the flat betting on a little % of your bank (1% or 0,5%) the problem are the time (did worth it?) and the book margin (have you calculate it ? 1 unit every 20 bets on the best books).

a questions: with your analisis had you see big momentary lose tread? or the winning are constantly in the time?

[Flight]

Randomly divide your data into a development/training data set and validation/test data set and start testing. The problem is most models depend on data up to a point in time, in other words, models gain accuracy with more data but should not be used to test data that was used to develop the model. So completely random division will not work. With a data set that large, you should still be able to come up with a logical way to divide into two sets that can be tested multiple times in different seasons. If you are random in your test samples, then you minimize the risk of mining error.

As far as market corrections go, you can compare your lines to market lines for yet another section of data and analyze your models error. If you find an edge and test with rigor, there is no reason to be cautious.

[Flight]

I'm sure someone better at the math side will hammer me for this, but its my opinion that seldom if ever do we reach the "long term". As you've alluded to, market adjustments and the constant evolution of both our own betting style and the market itself will render your older data irrelevant.

Also if you have enough data to conclude that you have been on the right track in the past - its exactly that- you were at that point in time doing enough to beat the market. It obviously doesn't guarantee that what your doing in the future will continue to beat the market. Thus I think its almost irrelevant to gather the sort of data required to figure out if you have been winning against the market.

I think in a lot of ways we just have to have faith in what were doing.

Wrong. Long term is a generalized term to describe lots of samples with variance built in. We use it to describe something with statistical significance, and is one of several required criteria to identifying a significant relationship between two variables. In other words, you can either be lucky in the short term, or have an edge in the long term. To think you never reach long term is a huge mistake because you are rejecting the importance of data, statistics, mean/stddev, samples, etc.

Have faith in what were doing = you are doomed to fail at gambling

If you cannot prove with math that you are not playing at -EV, then you cannot win. You will wind up "in the long term" losing money. The only thing that varies from player to player is how long it takes them to stop being lucky and just lose (regress to the mean of -4.5% actual value)

[roasthawg]

That's plenty of data.

[statnerds]

a questions: with your analisis had you see big momentary lose tread? or the winning are constantly in the time?

Of course a couple significant losing trends in there. Longest one I could identify was over 867 events. Oh how I wish they were constant and then I could time the market, know when to bet and when not to bet and make an absolute killing.

That 867 stretch would have erased ~48 Units off my BR.

[statnerds]

As far as market corrections go, you can compare your lines to market lines for yet another section of data and analyze your models error. If you find an edge and test with rigor, there is no reason to be cautious.

Good stuff Flight, thanks for the input.

There is always reason to be cautious. I am a conservative bettor by nature, but still take my shots when I feel it is justified. This data set and the approach pull on several pillars of my gambling belief.

Identify the opportunity

Proper money management

Beat the closing line.

Without the last one, this endeavor is doomed to fail, and I recognize that. Makes it more work, but over the course of 2,500 or 3,000 plays, years, it will be profitable.

Now I need to develop ideas on how to maximize my return.

[the_mathman]

Good stuff Flight, thanks for the input.

There is always reason to be cautious. I am a conservative bettor by nature, but still take my shots when I feel it is justified. This data set and the approach pull on several pillars of my gambling belief.

Identify the opportunity

Proper money management

Beat the closing line.
...

I like your approach!

[the_mathman]

Of course a couple significant losing trends in there. Longest one I could identify was over 867 events. Oh how I wish they were constant and then I could time the market, know when to bet and when not to bet and make an absolute killing.

That 867 stretch would have erased ~48 Units off my BR.

hum... dont understimate this information.
can you plan a money management that can resist at -100 units situation?

[Data]

Or would you only focus on criteria you felt would be unaffected by the evolution?

Yes. If the "system" ("model") is based on the evolving factors such as ATS records then it has no predictive value no matter how big the sample size is.

[statnerds]

Ouch, a system says you.

Your post made me ponder the lasting affects of market corrections. How many times does the market over-react to a short term result?

Then it would be like bell bottom jeans. Here

Then gone

Then here again.

Too simplistic?

[roasthawg]

Yes. If the "system" ("model") is based on the evolving factors such as ATS records then it has no predictive value no matter how big the sample size is.

Why? What's the difference between "evolving" factors and game stats? The game itself is just as likely to evolve as the market, no? Strategies, players, and rules change all the time.

[Data]

It is the difference between the changes we know about and can account for vs. the changes we do not know about.

[roasthawg]

It is the difference between the changes we know about and can account for vs. the changes we do not know about.

Tough to account for changes in playing style or coaching style. Even rule changes are tough to quantify since you're not gonna have much data initially.

Similarly, it's very tough to account for what the market is currently looking at. A large enough sample size should give you as good a chance as any at figuring out what the market will do going forward.

That's my take at least.

[Data]

Whatever is tough to account for is equally tough for the bettor and for the bookies. Some of the changes you mentioned have a little impact on the "global" stats anyway. On the other hand, a bettor may have an edge due to bookie not accounting for a certain factor F. There are two completely different takes on this. One bettor creates a model and makes a line for a given game, say, -5 while accounting for a factor F. The bookies' line is -3. This bettor knows that he has an advantage and keeps wining until the bookies catch up a few seasons later and start hanging -5 on a similar contest making our bettor realize that now he has no edge and there is no bet. Another bettor notices that betting the games with factor F produces a winning ATS record. He starts betting every game that meets that criteria. He does not know when to stop and will end up a loser due to the followers dilemma.

[icsky3]

Man sounds really good....Id like to see where this goes

[roasthawg]

Whatever is tough to account for is equally tough for the bettor and for the bookies. Some of the changes you mentioned have a little impact on the "global" stats anyway. On the other hand, a bettor may have an edge due to bookie not accounting for a certain factor F. There are two completely different takes on this. One bettor creates a model and makes a line for a given game, say, -5 while accounting for a factor F. The bookies' line is -3. This bettor knows that he has an advantage and keeps wining until the bookies catch up a few seasons later and start hanging -5 on a similar contest making our bettor realize that now he has no edge and there is no bet. Another bettor notices that betting the games with factor F produces a winning ATS record. He starts betting every game that meets that criteria. He does not know when to stop and will end up a loser due to the followers dilemma.

I guess the first question to answer is whether books set lines to:
1) be true 50/50 coinflips
2) attract even amounts of money on each side
3) attract more money on the losing side

I take it that you believe scenario #1 is what the bookies are shooting for?

[Data]

A mix between (1) and (2). They would want a pure (1) but have no way of making this happen. So, instead they settle for (2) with a caveat that mostly the sharp action is accounted for as this is the best approximation of (1). Don't know what (3) is.

Regardless, there is no rational argument to be made that the line will remain at -3 for years to come which is the requirement for the ATS jockey to be profitable.

[statnerds]

Don't want to stop the flow, but I have to say there are occasions where Books take a position on a game. Some might be accidental due to being slow to adjust lines. But I have no doubt that a book like Pinny or Bookmaker might take a position on a game and not necessarily 'sucker' bettors into playing the other side, but just make it more attractive. There has to be someone on this forum that watches lines all day and can attest to this fact, and state it with more eloquence than I just did.

What I need is a fellow bettor in every NFL city with a local that shades lines to the home team every week. Assuming locals elsewhere do the same.

[roasthawg]

A mix between (1) and (2). They would want a pure (1) but have no way of making this happen. So, instead they settle for (2) with a caveat that mostly the sharp action is accounted for as this is the best approximation of (1). Don't know what (3) is.

Regardless, there is no rational argument to be made that the line will remain at -3 for years to come which is the requirement for the ATS jockey to be profitable.

Well if the books are moving the lines based off of money and you believe the -3 will move to -5 then you must believe in the efficient market theory. I'm not so sure of that theory myself... in practice I don't see it happening.

Also, why wouldn't the ATS jockey also recognize that the -3 is now -5? Unless you're talking about system plays that don't account for the spread then he should be able to see that the line is now adjusted.

[statnerds]

Well if the books are moving the lines based off of money

Or do they?

[Data]

Well if the books are moving the lines based off of money and you believe the -3 will move to -5 then you must believe in the efficient market theory. I'm not so sure of that theory myself... in practice I don't see it happening.

First, I am talking about the market adjustment over a long period of time, not the intra-day line moves for a given contest. Second, the acknowledgment that there is a certain degree of efficiency in the market is completely different from the EMT.

Also, why wouldn't the ATS jockey also recognize that the -3 is now -5?

He has no way of doing so.

Unless you're talking about system plays that don't account for the spread then he should be able to see that the line is now adjusted.

I am talking about the "systems" that make ATS choices based on ATS records, ex. "play NFL home dogs".

[roasthawg]

I am talking about the "systems" that make ATS choices based on ATS records, ex. "play NFL home dogs".

Ah, it took me awhile to see that this is what you were talking about... I thought you were talking about different types of variables in a regression and distinguishing between them as evolving or not.

[Rich Boy]

Don't want to stop the flow, but I have to say there are occasions where Books take a position on a game. Some might be accidental due to being slow to adjust lines. But I have no doubt that a book like Pinny or Bookmaker might take a position on a game and not necessarily 'sucker' bettors into playing the other side, but just make it more attractive. There has to be someone on this forum that watches lines all day and can attest to this fact, and state it with more eloquence than I just did.

What I need is a fellow bettor in every NFL city with a local that shades lines to the home team every week. Assuming locals elsewhere do the same.

I watched the SBR interview with Pinnacle and they said they dont take sides on a game, they just like to take the initial sharp action and use that information to shape their lines. But the guy wasnt that helpful, could have been lying/trying to hide some information.

[Wrecktangle]

Firstly, it is almost impossible to have too much data. But unfortunately in some leagues such as the NFL, the rules and coaching factors relating to this change so frequently (yearly now) as to make earlier years fairly worthless. I have line data and boxscores from the 70's (my NFL data set goes back to 1970) and the predictive value related to today is not worthwhile. In College football (a data set going back to 1976) is better as the last major rule set was 1994 or so when the overtime rules were implemented.

But even in the massive CFB data set for some infrequently occurring factors, I haven't enough observations because of the type of modeling I do, since for some data types I set the maximum look back date to 1994.

There are no easy answers to this question. At lot of testing is involved and continuous observation of your results is necessary.

[statnerds]

RB - I don't trust any of them bastards to tell the truth. But if they are telling you sharp action dictates early line movement, sounds like beating the Pinny mover is viable. Which makes wonder if they only release lines to select clientele before posting them to the general public.

WT - I love you and your insights. Reading through your response made me think that it is the amount of data getting too large, it matters only if you are properly analyzing the data along with rule changes and coaching philosophies and a host of other things. So the data is not a stand alone factor or predictive force.

Thanks guys.

[RickySteve]

I want the last 3 minutes of my life back.

[statnerds]

I want the last 3 minutes of my life back.

What were you expecting to find based on the thread title?

[sharpcircle]

I want the last 3 minutes of my life back.

you read very fast

[Inkwell77]

to all the posts