A question for the math guys

[formula_2002]

A question for the math guys

I have what appears to be a wonderful correlation between my figures and final odds.
It looks something like this.
When the average figure is 1 the odds are 1
Average figure is 2 and the odds are 2 and so on and so on.
The roi for each of these incremental sets is about what you would expect for the odds, about, .84 and changing due to the long to short odds bias.

By running some comparisons, it is somewhat obvious that when the figure is 1 and the odds are 10, bingo, the roi goes above 1.

Question.
What is the formula to determine the optimal relationship between the figures and the odds for the maximum the roi.
Reference to a math site would be ok.

[Tom]

Not sure about a formula, but is sure likes a graph ( demand curve?) I saw in an old Economics book. could you express the relationship as an index - factor/odds, then plot roi against it? Your curve would max out at some point - which is what you need.

[formula_2002]

Not sure about a formula, but is sure likes a graph ( demand curve?) I saw in an old Economics book. could you express the relationship as an index - factor/odds, then plot roi against it? Your curve would max out at some point - which is what you need.
The first thought that came to mind was my econominc's teacher, a Mr Wolfson, drawing those curves onto a chalk board.

I tried that, it didn't work out. I think its because the facator is averaged.
remember, it's average 1 =odds when odds >=.5<=1.4
There is a small deviation in the average .

[rrbauer]

Need some clarification:

When the average figure is 1 the odds are 1

"average" odds are 1? And you have std deviation for
each fig set; and, each odds within fig set?


By running some comparisons, it is somewhat obvious that when the figure is 1 and the odds are 10, bingo, the roi goes above 1.

So the "expected" odds are 1 (based upon the fig) but the actual odds are 10? Do you have some sort of frequency distribution of actual odds/results by fig set?

[Robert Fischer]

What is the ROI of a given toteOdds y for a given average figure x ?

I think Tom is correct in that you plot this. It should be a bit of a curve. For example an horse going off between at 8-1 with an estimate odds line of 1, may be a great ROI... Yet, the rare horse going off at 30-1 with an estimate odds line of 1 may or may not show a profitable ROI....

You could find your profitable ranges... Work within them, and possibly write a correlation formula for those ranges ?

[formula_2002]

I have to do a deeper analysis, after which optimal ROI's may become more apparent.

But to be clear, each figure is averaged to an incremental odds analysis. The odds range >0<=.5, >.5<=1, >1<1.5 etc.. the standard stuff.



Now, Ive been extracting the distribution of each of the averaged factor figures.

I'll do a bit more work and take another look at the requirement for a formula.

Thanks

Joe M

[formula_2002]

Ok, here is the deal.
Example:
looking at odds >=1.and.<2-1
average factor is 930, with a .83 roi.

Now, when we look at factors, say, >=1100 the roi jumps to .91, and when we look at factors, say <900
the roi gets down to .75..

So what should be done is, establish relationships between very small increments of odds and factors, determine the win % at each relation.

Then you can start to determine an odds line based on factors and public odds.

Look at horse 1:
My factor is say 1100, the public has him at 1.5-1, my relationship has predetermined a .376 probability of winning.
Do that for every horse in the field and re-calculate the odds and play the overlays.

I'll see how that works out.. It sounds like a good plan.(?)

[formula_2002]

Hey, maybe I'm the math guy! ;)
I have determined the win % for each figure based on an incremental odds analysis.
example:
in the odds range >=3.5 <4.5 I get the following win%
figure WIN_PER
49. 0.077
57. 0.107
53. 0.110
47. 0.123
56. 0.131
50. 0.132
54. 0.141
51. 0.156
55. 0.204

I'm not happy with that distribution, so I wrote a program that makes some adjustments. The results are:
FACTOR adj win %
47. 0.09
49. 0.10
50. 0.11
51. 0.12
53. 0.13
54. 0.14
55. 0.15
56. 0.16
57. 0.17

curious to the results.

[PriceAnProbability]

Question.
What is the formula to determine the optimal relationship between the figures and the odds for the maximum the roi. Reference to a math site would be ok.

There is a fixed relationship between expected speed figure (probability) and odds distribution (price). I.e., if you expect a six-horse field to earn figures of 80, 82, 71, 64, 77, and 59, you can convert that to an odds line if you have figured out the formula. Since I prefer not to have to get a day job, I'll leave it to you to develop your own conversion equation.

Now, if you want a simple way of exploiting your ratings, just rank your horses in order of rating, and bet the first horse who is "off the chalk," i.e., where you rank the horse in a higher finish position than the program morning line (not the final odds).

[formula_2002]

There is a fixed relationship between expected speed figure (probability) and odds distribution (price). I.e., if you expect a six-horse field to earn figures of 80, 82, 71, 64, 77, and 59, you can convert that to an odds line if you have figured out the formula. Since I prefer not to have to get a day job, I'll leave it to you to develop your own conversion equation.

Now, if you want a simple way of exploiting your ratings, just rank your horses in order of rating, and bet the first horse who is "off the chalk," i.e., where you rank the horse in a higher finish position than the program morning line (not the final odds).

PriceAnProbability,
The wonder of computing power and comma delimited data is, in this case, no need to worry about formulas too much. But I will admit, this treak is giving me new cause to appreciate the calculus.
The volume of data I have, contains the information to establish the correlations. Setting up the data for analysis is a lot of grunt work.
Right now I know or could know within minuets, what I could expect of a Bris say 80 speed figure when the public odds are >=1.and.<1.5-1.
as a mater of fact, I just figured it out,.In 101 horses of the 120,000 horses in my testing data base (about 1/3 of all my data) fit that condition. The beast won 34% of the time.
Now I just test those conditions in another data base, also 120,000 horses. There the beast won 41% of the time in 129 races.
The formulas come into play when you try to establish the curves for the expected results base on too little incremental data . What happens when that figure 80 Bris speed is broken down by various class, surface,distance,track, etc...

Hey, this stuff keeps me away from the betting windows! ;)

[PriceAnProbability]

PriceAnProbability,
The wonder of computing power and comma delimited data is, in this case, no need to worry about formulas too much. But I will admit, this treak is giving me new cause to appreciate the calculus.
The volume of data I have, contains the information to establish the correlations. Setting up the data for analysis is a lot of grunt work.
Right now I know or could know within minuets, what I could expect of a Bris say 80 speed figure when the public odds are >=1.and.<1.5-1.
as a mater of fact, I just figured it out,.In 101 horses of the 120,000 horses in my testing data base (about 1/3 of all my data) fit that condition. The beast won 34% of the time.
Now I just test those conditions in another data base, also 120,000 horses. There the beast won 41% of the time in 129 races.
The formulas come into play when you try to establish the curves for the expected results base on too little incremental data . What happens when that figure 80 Bris speed is broken down by various class, surface,distance,track, etc...

Hey, this stuff keeps me away from the betting windows! ;)

I don't see how you can say what an "80" would do unless you know what it's up against. If you don't know how to convert figures to odds, just work by rank against the program ML rank and you'll get approximately the same thing.

I also don't believe in using large samples to derive this data, but rather trial-and-error with your ROI as the measuring stick. Much more efficient.

[formula_2002]

I don't see how you can say what an "80" would do unless you know what it's up against. If you don't know how to convert figures to odds, just work by rank against the program ML rank and you'll get approximately the same thing.

I also don't believe in using large samples to derive this data, but rather trial-and-error with your ROI as the measuring stick. Much more efficient.

You are right, I dont know what an 80 will do (although I know what it did).
So I look into it to find out what it MIGHT do. But if its a good figure, it should be harmonious with the other figures. Just what does that mean?
No too sure, so again, I look into it to find out.

[formula_2002]

I have a nice set of figures vs odds for the 120000 horse data base.

I did some manual testing of a few races.
There is enough of a difference between the final odds and the data base determined odds to make it interesting.

I'll test these figures against a live board tomorrow.

[formula_2002]

Just completed some testing against my nov 2006 and a few days of Dec 2006.
Results look good (profits that is).
Good enough for me to run the test against my entire 2006 data base.
Should be able to get it done today.

[PriceAnProbability]

Just completed some testing against my nov 2006 and a few days of Dec 2006.
Results look good (profits that is).
Good enough for me to run the test against my entire 2006 data base.
Should be able to get it done today.

You mean if you blindly bet horses who earn a given figure they are profitable?

Is there a real-world example of this?

[formula_2002]

You mean if you blindly bet horses who earn a given figure they are profitable?

Is there a real-world example of this?

"Blindly"?. A weeks work on what has to be on a data base costing over 10 g's is not "blindly" :(
Let me run some more results before I give examples.

[formula_2002]

I have determined the win % for a single factor at small incremental odds for several thousand horses over a three month period. The win % were then readjusted by a formula.

The win % were determined using data outside of the data base that was used for this test.



Here are some partial results for that test when final odds are <=10-1.

when the formulated win % was <=win% determined by the final track odds, 12832 plays returned a .81 roi

when the formulated win % was =win% determined by the final track odds, 4013 plays returned a .84 roi

when the formulated win % x .9 was >=win% determined by the final track odds, 1803 plays returned a .89 roi

when the formulated win % x .8 was >=win% determined by the final track odds, 486 plays returned a 1.02 roi

when the formulated win % x .7 was >=win% determined by the final track odds, 245 plays returned a 1.03 roi

[PriceAnProbability]

I have determined the win % for a single factor at small incremental odds for several thousand horses over a three month period. The win % were then readjusted by a formula.

The win % were determined using data outside of the data base that was used for this test.



Here are some partial results for that test when final odds are <=10-1.

when the formulated win % was <=win% determined by the final track odds, 12832 plays returned a .81 roi

when the formulated win % was =win% determined by the final track odds, 4013 plays returned a .84 roi

when the formulated win % x .9 was >=win% determined by the final track odds, 1803 plays returned a .89 roi

when the formulated win % x .8 was >=win% determined by the final track odds, 486 plays returned a 1.02 roi

when the formulated win % x .7 was >=win% determined by the final track odds, 245 plays returned a 1.03 roi

What is the "forumulated win %?"

So far you haven't even identified what type of horses you are studying or how to bet them.

Like if I asked you what 2+2 was and you replied "Jello."

[formula_2002]

What is the "forumulated win %?"

Too varied to answer here just yet. This is a along process. Lots of things to look into. I'm just sharing as I go along.

So far you haven't even identified what type of horses you are studying or how to bet them.


Horses of all ages,sex,race types,distance,surface,etc. It's all one group.
I would not bet on any of them just yet.


Like if I asked you what 2+2 was and you replied "Jello"]
No. I'lll say I dont have enough information ;)

[classhandicapper]

Too varied to answer here just yet. This is a along process. Lots of things to look into. I'm just sharing as I go along.

Horses of all ages,sex,race types,distance,surface,etc. It's all one group.
I would not bet on any of them just yet.

No. I'lll say I dont have enough information ;)

I think if you start searching through "logical" sub groups of horses you will be able to increase the ROI.

[formula_2002]

I think if you start searching through "logical" sub groups of horses you will be able to increase the ROI.
One of the difficulties in isolating on sub groups is, you start to run out of data.
Take 4 catagories of class, and two of distance and you have cut you data base into 8 sets. Now instead of looking at 28,000 races, I'm looking at 8 sets of perhaps 3,500 .
Break that down further into different tracks and take-outs and before you know it, you are looking at several hundred races. break that down further into another factor and there you are,enough data to write onto a stamp. ;)
I'm not saying that approach is wrong, its just not pratical for me.

[classhandicapper]

One of the difficulties in isolating on sub groups is, you start to run out of data.
Take 4 catagories of class, and two of distance and you have cut you data base into 8 sets. Now instead of looking at 28,000 races, I'm looking at 8 sets of perhaps 3,500 .
Break that down further into different tracks and take-outs and before you know it, you are looking at several hundred races. break that down further into another factor and there you are,enough data to write onto a stamp. ;)
I'm not saying that approach is wrong, its just not pratical for me.

I understand your concerns.

Sometimes instead of breaking things down into units like that, you can weed out certain horses and increase your ROI without hurting the sample size significantly and while also making handicapping sense.

For example, if part of what you are looking at is speed figures, you might want to weed out horses that earned good figures on turf if today's race is dirt. You could also consider pedigree as part of that.

As you remove groups of negative horses, the remainder keeps getting better and better.

[PriceAnProbability]

My question again here is:

Exactly what type of horse is he studying as a wager?

What is the criteria for betting the horse?

I still don't see any.

[PlanB]

Formula, long ago, before PA banned me, I typed my best post. It was in English & I kinda speculated on a DB size sufficient to grapple with the hurdles. In my very simple math I calculated that very often you need ~600K races just to start. Now, of course, that's using standard math stuff; more modern stuff could reduce that to ~300K races. BUT BUT these races must be random. So, the pool must be in excess of 700K races. ummm, is there a way out? YES. YES.

[Tom]

So you obvioulsy bet after you retire, eh?
700K races?!?

[formula_2002]

My question again here is:

Exactly what type of horse is he studying as a wager?

What is the criteria for betting the horse?

I still don't see any.
Here is an example from my aug 2006 data base, which was not included in my previous report.
08/04/2006 at AP, race 1, maiden sprint, horse #'s 3 and 7 each had a calulated win % 20 above the final odds.
The 7 horse won and paid $4.80.

In that month there were 240 plays(20% calculated overlay, odds<10-1) that returned a $490 for a 2% profit.
This is very much inline with the previous report.

[formula_2002]

Formula, long ago, before PA banned me, I typed my best post. It was in English & I kinda speculated on a DB size sufficient to grapple with the hurdles. In my very simple math I calculated that very often you need ~600K races just to start. Now, of course, that's using standard math stuff; more modern stuff could reduce that to ~300K races. BUT BUT these races must be random. So, the pool must be in excess of 700K races. ummm, is there a way out? YES. YES.

Actually, I recall going at it with a bunch of guys here when I said it would be impossible to gather enought data to prove any winning system.
But I have to set that thought aside, else what would I do with my "free" time ;)

I just may write a protected program and put it up on my web site and you can test my formula using your data..Hows that sound?

[PlanB]

Actually, I recall going at it with a bunch of guys here when I said it would be impossible to gather enought data to prove any winning system.
But I have to set that thought aside, else what would I do with my "free" time ;)

LOL. if you need me to land a low punch or two call me. It's like the # Xmas days left to shop, it creeps on you, unexpected. "Many bad buys follow," CONFUSIUS say. ps, "man who fart in church sit in own pew." hehe.

[PriceAnProbability]

Here is an example from my aug 2006 data base, which was not included in my previous report.
08/04/2006 at AP, race 1, maiden sprint, horse #'s 3 and 7 each had a calulated win % 20 above the final odds.
The 7 horse won and paid $4.80.

In that month there were 240 plays(20% calculated overlay, odds<10-1) that returned a $490 for a 2% profit.
This is very much inline with the previous report.

Unless you mention how to calculate the win percentage, there's no way to analyze the results of your study, since we don't know what you are studying.

[formula_2002]

Unless you mention how to calculate the win percentage, there's no way to analyze the results of your study, since we don't know what you are studying.
Well, thats not going to happen.
If I think I have something of value, I'll write a downloadable program that will allow people to test their own data files.

[PriceAnProbability]

Well, thats not going to happen.
If I think I have something of value, I'll write a downloadable program that will allow people to test their own data files.

Well then you're just talking to yourself if you don't share the underlying formula, since no one knows what it is you're studying.

[PlanB]

Formula shares his knowledge more than 99% of you, particularly YOU system sellers, purveyors of data, raw & worked over. Grow up. A few of you here are relentless salesmen, forever selling, and NEVER PICKING ANY RACES. In a way, you're like the drug companies, telling us how great their latest drug is while all takers drop dead from side effects. ps: I also pay for a data service & I love it but you don't see the moronic claims from it; I just lack the savvy on how to use it, but I am improving.

[PriceAnProbability]

Formula shares his knowledge more than 99% of you, particularly YOU system sellers, purveyors of data, raw & worked over. Grow up. A few of you here are relentless salesmen, forever selling, and NEVER PICKING ANY RACES. In a way, you're like the drug companies, telling us how great their latest drug is while all takers drop dead from side effects. ps: I also pay for a data service & I love it but you don't see the moronic claims from it; I just lack the savvy on how to use it, but I am improving.

I'm not talking about sharing picks or information, but rather what you are trying to claim is a profitable angle, since you asked for feedback. My point is that you gave nothing for one to work with to give you feedback on, so there's nothing to evaluate.

[formula_2002]

I'm not talking about sharing picks or information, but rather what you are trying to claim is a profitable angle, since you asked for feedback. My point is that you gave nothing for one to work with to give you feedback on, so there's nothing to evaluate.

P&P
the following is from post #1
Question.
What is the formula to determine the optimal relationship between the figures and the odds for the maximum the roi.
Reference to a math site would be ok.
Did I miss your answer?
I claim noting as a "profitable angle".

[PriceAnProbability]

P&P
the following is from post #1

Did I miss your answer?
I claim noting as a "profitable angle".

"optimal relationship" implies that winning is your goal.

"The figures" and "the morning line" means "how do you assign probability to a horse based on its figure." But then you said you studied all horses with a figure of X (something like 80), which serves no purpose, since that doesn't account for what the "80" is running agains, or even begin to explain why "80" would be a more profitable figure than "81" or "79" etc.

[formula_2002]

"optimal relationship" implies that winning is your goal.

"The figures" and "the morning line" means "how do you assign probability to a horse based on its figure." But then you said you studied all horses with a figure of X (something like 80), which serves no purpose, since that doesn't account for what the "80" is running agains, or even begin to explain why "80" would be a more profitable figure than "81" or "79" etc.

If that is how you see it, then take it to the next level.
Determine the formula in which the roi can be developed (and maximized) with respect to x and it's relationship to the other "x's" its running against.
roi..

[PriceAnProbability]

If that is how you see it, then take it to the next level.
Determine the formula in which the roi can be developed (and maximized) with respect to x and it's relationship to the other "x's" its running against.
roi..

You mean the formula for how to convert a set of ratings into an accurate morning line.

Already did that with the Beyers. Give me a set of expected Beyer figures for a race, and I can convert it into an accurate morning line.

Now all you need to do is guess the horse's Beyer figure today and you're set!

[traynor]

One of the difficulties in isolating on sub groups is, you start to run out of data.
Take 4 catagories of class, and two of distance and you have cut you data base into 8 sets. Now instead of looking at 28,000 races, I'm looking at 8 sets of perhaps 3,500 .
Break that down further into different tracks and take-outs and before you know it, you are looking at several hundred races. break that down further into another factor and there you are,enough data to write onto a stamp. ;)
I'm not saying that approach is wrong, its just not pratical for me.

Another difficulty in extrapolating future probabilities from historical data (regardless of sample size) is that you never wagered in those races. That is, there is no way to determine the distribution of the mutuel pools in future races based on historical evidence, other than in broad generalities.

Suggestion: Once you get your algorithm relatively stable, run a "real-time" test on current races. You still won't be able to measure the effect of wagering (unless you bet) but you can learn a lot from the actual odds available on the entries you would use, relative to the size of the mutuel pools. If you have enough slack to absorb decent wagers, you may really have something.

Conversely, a 3% or 4% edge can magically disappear when fairly modest (additional) wagers are made on those selections.
Good Luck

[formula_2002]

You mean the formula for how to convert a set of ratings into an accurate morning line.

Already did that with the Beyers.

Thats a good start..want to share it?

[PriceAnProbability]

Thats a good start..want to share it?

Share the method for estimating today's speed figure? No.

The method for converting that estimate into a morning line? Maybe.

[formula_2002]

Share the method for estimating today's speed figure? No.

The method for converting that estimate into a morning line? Maybe.

That would do for me..

[PriceAnProbability]

That would do for me..

Damn, with demand like that for this info I should write a book!

Anyone else interested in this formula?

[Robert Fischer]

hey , all the smart-ass stuff aside, I hold in high regard anyone who is seeking knowledge and doing some of the ground-work themselves.
Thank you Formula for sharing as you are working on this.

[PriceAnProbability]

hey , all the smart-ass stuff aside, I hold in high regard anyone who is seeking knowledge and doing some of the ground-work themselves.
Thank you Formula for sharing as you are working on this.

Most of this theory was developed about a decade ago when bookies began offering money lines on NFL games. They found that each pointspread had a money-line conversion, and the two are usually correlated. This in turn was based on the method used to generate Canadian-style odds on NHL games.

The same would be true for beaten lengths, or speed figure expressions of same.

[formula_2002]

The same would be true for beaten lengths, or speed figure expressions of same.

So cool!! and the formula is...

A good bookie follows , I would think, the MONEY.
He should give a rat's A.. about anything else.

[formula_2002]

hey , all the smart-ass stuff aside, I hold in high regard anyone who is seeking knowledge and doing some of the ground-work themselves.
Thank you Formula for sharing as you are working on this.

RF, if it were within my power to share anything really worth the while (profits that is) I would, at this point in my life, do so willingly.

This is a brutal game.
Although my "Beitandforgetit picks made a nice profit when >=5-1 in November (see selections forum), I could not say it would repeat again.

Joe M

[PriceAnProbability]

So cool!! and the formula is...


...apparently more valuable than I had thought.