How do you approach ATS/Total margins?

[pro-style]

For NBA I have found a lot of trends that may be hitting in the 54.5% area but may have a very small margin ATS or against the total. I've found some trends with 500+ game samples that hit in the same range in terms of percents, but one may have a margin of 0.2 and the other may have a margin of 1.3.

How should this factor in to the confidence that the trend is actually more than just a fluke. Keep in mind that these are pretty big sample sizes...

[pro-style]

Let me give concrete example:

Trend X is 368-300 (55.01%) and has a margin of 0.7.
Trend Y is 345-300 (53.48%) and has a margin of 1.4.

How do these compare in terms of calculated edge.

Let's assume the sport is NBA so the point value is closer to linear, unlike football.

[pro-style]

Anyone ?

[hutennis]

Calculated edge? Are you talking past edge? Or future edge?

[pro-style]

Calculated edge? Are you talking past edge? Or future edge?

Future edge.

[MonkeyF0cker]

Calculated edge? Are you talking past edge? Or future edge?

Neither isn't a choice?

[hutennis]

Future edge.

What makes you believe that old patterns you have stumbled upon will deliver any kind of edge in a future?

[pro-style]

What makes you believe that old patterns you have stumbled upon will deliver any kind of edge in a future?

The two main things I look for:

1) Very little or no regression back to the 50/50 point. I usually graph these results and see if there is any regression back to the 50/50 mark. I take this as an indicator that the market has not adjusted for this scenario.

2) It has a very low cumulative probability to be a "fluke" based on looking at this as a binomial distribution.

Now I am trying to figure out how to factor in margins when points are close to a linear scale.

I have a limited exposure to advanced statistics (other than a few engineering level courses), so if you see any flaws in my logic, please discuss.

EDIT:

Also, another thing I kind of look at is how random a trend's parameters are. If its stuff that doesn't even seem like it should be relevant, I look at that as more of a chance of being a fluke.

For example: If the parameters are logical like... "teams off X point comeback win and now they are facing a 'bad' team" and the results come back as 200-250 with an ATS margin of 1-2 points I look at it as more likely to have true value in the future.

[Wrecktangle]

Calc z-scores on each "trend" using binomial dist. with default being .5

Z-scores will give you some idea if your "trend" is not a fluke, especially then above 2 or below -2, but bear in mind leagues change from year to year, especially NBA this year due to strike.

[pro-style]

Calc z-scores on each "trend" using binomial dist. with default being .5

Z-scores will give you some idea if your dist is not a fluke, especially then above 2 or below -2, but bear in mind leagues change from year to year, especially NBA this year due to strike.

Ok thanks, I'll start looking into that. I've seen that mentioned before, but didn't cross my mind at all.

[MonkeyF0cker]

Uhh. You're already doing that.

2) It has a very low cumulative probability to be a "fluke" based on looking at this as a binomial distribution.

[pro-style]

Uhh. You're already doing that.

Yea, I realized this later in the day. I just didn't know how to translated to a z-score.

I have got some pretty high z-scores on some trends. One even hit a little higher than 3. Been having recent success with these so we'll see if it continues.

[hutennis]

2 things.

First.
For all this z-score/binomial distribution business to be valid you need to be absolutely sure that subject you explore is governed by bell curve as completely as roulette wheel is. I have no idea how that assumption can be made.

Second.
The chance that pattern you are looking at is already discovered, priced in and reflected in odds, thus making your effort a complete waste of time, is overwhelming.
You can lose a lot of money in a process of confirming this empirically, or you can use common sense and very safely assume that finding an undiscovered long term profitable trend in a monstrous size market like NBA with millions upon millions of participants (some of whom are very sophisticated) looking at every aspect of it through the microscope is as likely as finding cold fusion.

Hope this helps.

[pro-style]

2 things.

First.
For all this z-score/binomial distribution business to be valid you need to be absolutely sure that subject you explore is governed by bell curve as completely as roulette wheel is. I have no idea how that assumption can be made.

Second.
The chance that pattern you are looking at is already discovered, priced in and reflected in odds, thus making your effort a complete waste of time, is overwhelming.
You can lose a lot of money in a process of confirming this empirically, or you can use common sense and very safely assume that finding an undiscovered long term profitable trend in a monstrous size market like NBA with millions upon millions of participants (some of whom are very sophisticated) looking at every aspect of it through the microscope is as likely as finding cold fusion.

Hope this helps.

I don't have much time right now, but I disagree with some of the stuff you said in your second portion, I'll get back to that later.

As for number 1, why can you not assume that it falls under a bell curve?

What if I were to randomly flip a coin and let that be the decision on whether I took team A or team B. Wouldn't this fall under a bell curve? How is this really any different? I'm determining the chance that given a set of past results, the outcome is just a fluke. Why is comparing this to a binomial distribution incorrect?

[hutennis]

As for number 1, why can you not assume that it falls under a bell curve?

For the same reason it's correct to assume that CLT does not work in stock market, or real estate, or publishing among others. Every time when it is not just Nature that rules unconditionally, but
when there is a chance for human nature to run wild (and I dont see why not in sports) bell curve is a huge suspect.

And yes, flipping coin is the great way to guaranty that you can never be beat in which drop runs faster contest, but I dont think you can be as much confident that you'll break even on next 1000 over/unders even though previose 1000 games have a perfect record of 500/500.
Too many ever changing unknown variables are in play.

[Wrecktangle]

For the same reason it's correct to assume that CLT does not work in stock market, or real estate, or publishing among others. Every time when it is not just Nature that rules unconditionally, but
when there is a chance for human nature to run wild (and I dont see why not in sports) bell curve is a huge suspect.

And yes, flipping coin is the great way to guaranty that you can never be beat in which drop runs faster contest, but I dont think you can be as much confident that you'll break even on next 1000 over/unders even though previose 1000 games have a perfect record of 500/500.
Too many ever changing unknown variables are in play.

As I said, leagues change from year to year. NBA has changed quite a bit this year. No guarantees in this biz.

BOL

[RickySteve]

As I said, leagues change from year to year. NBA has changed quite a bit this year. No guarantees in this biz.

BOL

You are a sage.