Limiting our betting choices to the two top Bris Prime Power horses

[DeltaLover]

Let's assume that we limit our betting choices to the top two bris prime power horses of the race and we have discovered a way to detect with 100% accuracy, which of the two will beat the other one.

Betting on this horse will show a 1.7 ROI

This assumption is of course impossible to achieve but it raises the question of how frequently our assessment needs to be correct, to guarantee a positive ROI or to just break even.

In other words, selecting our bet between the top two BPP, how many times we need to be correct, to have a fair game with a 1.0 ROI ?

What the optimal rate of first vs second choice should be?

Can you suggest of related methodologies that can be used for this purpose?

[Flysofree]

I don't understand your question..... Are you asking what should be the odds of both horses or one horse??? :confused:

[DeltaLover]

I don't understand your question..... Are you asking what should be the odds of both horses or one horse??? :confused:

I do not think I have mentioned odds anywhere in my posting.

The problem is the following:

(1) We restrict our selection to one of the top two BPP horses.

(2) We bet our selection for the same amount every time

The question is the following:

What percentage of the races we need to be correct in our selection between the top two horses in order to break even and start making a profit?

Is something like this possible at all?

[thaskalos]

Let's assume that we limit our betting choices to the top two bris prime power horses of the race and we have discovered a way to detect with 100% accuracy, which of the two will beat the other one.

Betting on this horse will show a 1.7 ROI

This assumption is of course impossible to achieve but it raises the question of how frequently our assessment needs to be correct, to guarantee a positive ROI or to just break even.

In other words, selecting our bet between the top two BPP, how many times we need to be correct, to have a fair game with a 1.0 ROI ?

What the optimal rate of first vs second choice should be?

Can you suggest of related methodologies that can be used for this purpose?

Have we finally run out of REAL topics...and we are now venturing into "impossible assumptions"? :)

[DeltaLover]

Have we finally run out of REAL topics...and we are now venturing into "impossible assumptions"? :)

The real topic is following Thask ! :)

[thaskalos]

The real topic is following Thask ! :)

Alright then! Let's get with it! :) :ThmbUp:

[Flysofree]

I do not think I have mentioned odds anywhere in my posting.

The problem is the following:

(1) We restrict our selection to one of the top two BPP horses.

(2) We bet our selection for the same amount every time

The question is the following:

What percentage of the races we need to be correct in our selection between the top two horses in order to break even and start making a profit?

Is something like this possible at all?


Yes, you need to be right in 50% of your races (if the winner pays 1-1) to break even.
Check other odds to see how often to break even.
This is betting on one horse...
To bet on two horses just find a dutching chart... Google the words betting dutch in horse racing.

[ReplayRandall]

I do not think I have mentioned odds anywhere in my posting.

The problem is the following:

(1) We restrict our selection to one of the top two BPP horses.

(2) We bet our selection for the same amount every time

The question is the following:

What percentage of the races we need to be correct in our selection between the top two horses in order to break even and start making a profit?

Is something like this possible at all?


Delta, you are in the realm of possibility, but must first change rule (1) to this: #1- We must use our top two BPP horses. Secondly, the odds on these horses must be dutched in proper ratio, as to achieve a doubling of the amount invested, thus needing a 50% strike rate to begin showing profit. At first glance it appears a player would be passing at least 60% of all races to have a possibility of success. This is just a starting point, but I believe you're on the right track (no pun intended).

[DeltaLover]

Yes, you need to be right in 50% of your races (if the winner pays 1-1) to break even.
Check other odds to see how often to break even.
This is betting on one horse...
To bet on two horses just find a dutching chart... Google the words betting dutch in horse racing.

My question, has to do with been right in selecting between the top two horses (NOT BEEN RIGHT ABOUT WHO IS GOING TO WIN THE RACE)..

The given is that having a fictitious method of always selecting the right horse (between 1st - 2nd) or in other words been 100% RIGHT, our ROI is going to be 1.7...

[DeltaLover]

Delta, you are in the realm of possibility, but must first change rule (1) to this: #1- We must use our top two BPP horses. Secondly, the odds on these horses must be dutched in proper ratio, as to achieve a doubling of the amount invested, thus needing a 50% strike rate to begin showing profit. At first glance it appears a player would be passing at least 60% of all races to have a possibility of success. This is just a starting point, but I believe you're on the right track (no pun intended).


I am not referring to dutching. Just bet a single horse every time

[Flysofree]

Must be something in the water.... :D

[ReplayRandall]

I am not referring to dutching. Just bet a single horse every time

I saw the title of thread-Limiting our betting choices to the two top Bris Prime Power horses and thought of a refinement starting point......my bad.

[cj]

Let's assume that we limit our betting choices to the top two bris prime power horses of the race and we have discovered a way to detect with 100% accuracy, which of the two will beat the other one.

Betting on this horse will show a 1.7 ROI

This assumption is of course impossible to achieve but it raises the question of how frequently our assessment needs to be correct, to guarantee a positive ROI or to just break even.

In other words, selecting our bet between the top two BPP, how many times we need to be correct, to have a fair game with a 1.0 ROI ?

What the optimal rate of first vs second choice should be?

Can you suggest of related methodologies that can be used for this purpose?

I would say it doesn't matter how often your assessment is correct if you don't know how often the horse wins when you are correct.

[DeltaLover]

I would say it doesn't matter how often your assessment is correct if you don't know how often the horse wins when you are correct.

The winning frequency is already part of the problem. Remember that we are betting top two bris prime power who win with a specific frequency.

[Overlay]

You say that betting on the correct horse with 100% accuracy will show a 1.7 ROI. That translates to an average, race-in/race-out mutuel of $3.40 on a $2.00 wager. Based on that, wouldn't the break-even point for selecting the winner out of the top two horses be 59 (rounded up from 58.82353) times out of 100, or 59% of the time? (For example, over the course of 100 races, you'd bet $200. At an average mutuel of $3.40, you'd have to hit 59 (59 x 3.40 = $200.60) winners in order to at least break even.)

[DeltaLover]

You say that betting on the correct horse with 100% accuracy will show a 1.7 ROI. That translates to an average, race-in/race-out mutuel of 3.40. Based on that, wouldn't the break-even point for selecting the winner out of the top two horses be 59 (rounded up from 58.82353) times out of 100, or 59% of the time? (For example, over the course of 100 races, you'd bet $200. At an average mutuel of $3.40, you'd have to hit 59 (59 x 3.40 = $200.60) winners in order to at least break even.)

Perfect :ThmbUp:

This is absoluetly correct and can be proven empirically by running a simulator.

Now, can this information be any helpfull for betting reasons?

[Some_One]

Perfect :ThmbUp:

This is absoluetly correct and can be proven empirically by running a simulator.

Now, can this information be any helpfull for betting reasons?

Don't you have to factor in that the winner of this two horse race won't matter if the two horses selected run 6th and 7th?

[DeltaLover]

Don't you have to factor in that the winner of this two horse race won't matter if the two horses selected run 6th and 7th?

As I said before, this is already taken care by the fact that we select top 2 pp who win approximately 55% (not exactly sure for this figure but it is very close). Obviously whenever any other horse is winning we are losing a unit.

[davew]

I don't follow you - is this a make believe dream?

you limit bet to either top 2 BPP - you know which of these 2 will place higher so bet on that one - you still will have less than 50% winners, unless of course you also know which races the winner will not be in top 2 BPP


In all seriousness, I have wanted to use BPP to make a quick and dirty probability line for races where horses have established records- but have ran into a brain block - should I do std dev and use number compared to a chart with number of entrants that race?

[Light]

top 2 pp who win approximately 55%

You are forgetting field size. Maybe the top 2 bpp can win 55% in 5 horse fields, but that percentage will naturally decrease as field size increases.

[davew]

You are forgetting field size. Maybe the top 2 bpp can win 55% in 5 horse fields, but that percentage will naturally decrease as field size increases.

also how much higher the top horse is than second
Bris says top in dirt win 31%
http://www.brisnet.com/cgi-bin/static.cgi?page=primepower

[DeltaLover]

You are forgetting field size. Maybe the top 2 bpp can win 55% in 5 horse fields, but that percentage will naturally decrease as field size increases.

As I said before, I do not consider anything else except BPP. For this purposes of this exercise i do not care about winning percentage at all, the only thing I measure is the Profit&Loss.

[DeltaLover]

I don't follow you - is this a make believe dream?
you limit bet to either top 2 BPP - you know which of these 2 will place higher so bet on that one - you still will have less than 50% winners, unless of course you also know which races the winner will not be in top 2 BPP


Can you prove the highlighted text? ( NOTE: I DO NOT IMPLY THAT WHAT YOU SAY HERE IS WRONG)

Also, again I do not care about winning percentage, I only consider P&L




In all seriousness, I have wanted to use BPP to make a quick and dirty probability line for races where horses have established records- but have ran into a brain block - should I do std dev and use number compared to a chart with number of entrants that race?

I do not try to extract any kind of a probability line in this particular exercise.

[ReplayRandall]

Perfect :ThmbUp:

This is absoluetly correct and can be proven empirically by running a simulator.

Now, can this information be any helpful for betting reasons?


Did you get around to answering the question or did someone else answer it?

[DeltaLover]

Did you get around to answering the question or did someone else answer it?

What is your take?

[davew]

Can you prove the highlighted text? ( NOTE: I DO NOT IMPLY THAT WHAT YOU SAY HERE IS WRONG

Nope, not without data from BRIS - taken from their site (but it is also related to field size 31% top winner, ??% second and I am guessing less than 20%)->



Win up to 55% of the Time!
So, just how good is Prime Power? Take a gander at these results: based on a lengthy study of tens of thousands of races, Prime Power hits 31% top winners! That's sounds great, but when the study is broken down further and we focus strictly on dirt races, the results get even more interesting:

Top Prime Power horses by three full points (3.0+) or better score 39% of the time!
Top Prime Power horses by six full points (6.0+) or better score 46% of the time!
Top Prime Power horses by ten full points (10.0+) or better score 55% of the time!

[ReplayRandall]

What is your take?


If Overlay's figures are correct as you say of 59% winners, and with 100% accuracy you can pick the winner out of the top two BPP's, and if doing so you will attain a ROI of 1.7, then for betting purposes you would need an average price of $5.762, or rounded up to $5.80 to meet your objective.

[Dave Schwartz]

Interesting conversation.

... then for betting purposes you would need an average price of $5.762, or rounded up to $5.80 to meet your objective

Replay,

"Needing $5.80" does not actually solve the problem because, logically, as the price goes up (from average) the hit rate goes down.

I think I know where Delta L is going with this, but will hold back comment for now.

[ReplayRandall]

Interesting conversation.



Replay,

"Needing $5.80" does not actually solve the problem because, logically, as the price goes up (from average) the hit rate goes down.

I think I know where Delta L is going with this, but will hold back comment for now.


It seems that some component/factor is missing.........Besides my missing brain....

[cj]

The winning frequency is already part of the problem. Remember that we are betting top two bris prime power who win with a specific frequency.

OK, didn't realize that.

[Robert Goren]

I think it is highly unlikely that you can come with a method to make betting the top choice and probably not most second choices show a profit. I would think despite the high win% ,the ROI is so low you be better off just tossing them if you must look at them at all. You are probably better off not even looking them. There is a math reason for not looking, but I screw it up every time I try to explain it. Something about a cat in box.

[DeltaLover]

If Overlay's figures are correct as you say of 59% winners, and with 100% accuracy you can pick the winner out of the top two BPP's, and if doing so you will attain a ROI of 1.7, then for betting purposes you would need an average price of $5.762, or rounded up to $5.80 to meet your objective.

When we say 59% we do not mean overall winners.

What we mean is that we will be correct 59% in our assesment that our selected horse will beat the other (for any position, regardelss if it won the race or not)

[DeltaLover]

I think it is highly unlikely that you can come with a method to make betting the top choice and probably not most second choices show a profit. I would think despite the high win% ,the ROI is so low you be better off just tossing them if you must look at them at all. You are probably better off not even looking them. There is a math reason for not looking, but I screw it up every time I try to explain it. Something about a cat in box.


So far in this thread, we have not tried to implement the hows of this method (which as you say, can very well be impossible to discover).

All we have done, is to define the theoretical break-even point of a model that selects between the top two bpp selections. In other words, we now know, that if we need to limit our selections to these two horses, we need to be correct X amount of times, to start making a profit.

[HUSKER55]

Nope, not without data from BRIS - taken from their site (but it is also related to field size 31% top winner, ??% second and I am guessing less than 20%)->



Win up to 55% of the Time!
So, just how good is Prime Power? Take a gander at these results: based on a lengthy study of tens of thousands of races, Prime Power hits 31% top winners! That's sounds great, but when the study is broken down further and we focus strictly on dirt races, the results get even more interesting:

Top Prime Power horses by three full points (3.0+) or better score 39% of the time!
Top Prime Power horses by six full points (6.0+) or better score 46% of the time!
Top Prime Power horses by ten full points (10.0+) or better score 55% of the time!



Since the win% will float per race, per card, per track, how do you come up with a rate.

How many races per card are 3+, 6+, 10+, and etc...The higher the spread the worse the payouts. Public beats it down.

I am assuming that <3points the bris figs are no better than the public at 33%. Then the top prime becomes 6% better than the public (3 to 6 points)?

So in 100 races you pick up 6 races over the public that did not pick the top prime power?

Wouldn't you have to base your approach on those races the public got wrong?

The beeper tells me my coffee is done...

I will be interested in your results.
:)

[DeltaLover]

Nope, not without data from BRIS - taken from their site (but it is also related to field size 31% top winner, ??% second and I am guessing less than 20%)->



Win up to 55% of the Time!
So, just how good is Prime Power? Take a gander at these results: based on a lengthy study of tens of thousands of races, Prime Power hits 31% top winners! That's sounds great, but when the study is broken down further and we focus strictly on dirt races, the results get even more interesting:

Top Prime Power horses by three full points (3.0+) or better score 39% of the time!
Top Prime Power horses by six full points (6.0+) or better score 46% of the time!
Top Prime Power horses by ten full points (10.0+) or better score 55% of the time!

This is not necessary for this thread, but since the winning frequencies of BPP came up a couple of times, this table is presenting them:


http://www.themindofagambler.com/bpp-winning-freq.jpeg

[davew]

Perfect :ThmbUp:

This is absoluetly correct and can be proven empirically by running a simulator.

Now, can this information be any helpfull for betting reasons?

Overlays data is with 59% winners, which is incorrect with your new data

With your 14890 race sample, the top 2 rated win roughly 52%, which you can predict 59% of the time narrowing down to 1 horse.

That gives you a little under 31% winners, which need to average $11.20 to get your goal ROI. If you start limiting bets to 5/1+, your win % will no longer be 31%.

[HUSKER55]

Thanks for the table Delta Lover

[DeltaLover]

Overlays data is with 59% winners, which is incorrect with your new data

With your 14890 race sample, the top 2 rated win roughly 52%, which you can predict 59% of the time narrowing down to 1 horse.

That gives you a little under 31% winners, which need to average $11.20 to get your goal ROI. If you start limiting bets to 5/1+, your win % will no longer be 31%.

Think it a little deeper.. We never said that we can win 59% of the time but something completely different... You confuse the win rating with the times we are correct when selecting between the top 2! It is not the same!

[dansan]

Why does have to win why not place at a price

[Overlay]

You say that betting on the correct horse with 100% accuracy will show a 1.7 ROI. That translates to an average, race-in/race-out mutuel of $3.40 on a $2.00 wager. Based on that, wouldn't the break-even point for selecting the winner out of the top two horses be 59 (rounded up from 58.82353) times out of 100, or 59% of the time? (For example, over the course of 100 races, you'd bet $200. At an average mutuel of $3.40, you'd have to hit 59 (59 x 3.40 = $200.60) winners in order to at least break even.)
Overlays data is with 59% winners, which is incorrect with your new data

With your 14890 race sample, the top 2 rated win roughly 52%, which you can predict 59% of the time narrowing down to 1 horse.

That gives you a little under 31% winners, which need to average $11.20 to get your goal ROI. If you start limiting bets to 5/1+, your win % will no longer be 31%.
Think it a little deeper.. We never said that we can win 59% of the time but something completely different... You confuse the win rating with the times we are correct when selecting between the top 2! It is not the same!
I may have contributed to the confusion by the way that I worded my earlier post. When I said "selecting the winner out of the top two horses", I was not referring to selecting the overall winner of the race. I instead meant (as DeltaLover indicated) selecting which one of the top two horses will finish ahead of the other, even if neither of the two horses actually finishes first in the race. If you could do that 100% of the time, the horse that is selected will actually win the race often enough to produce an overall ROI of 1.70.

In order to at least break even (i.e, an ROI of 1.00), you would have to correctly select the horse out of the top two prime power ratings that will finish in front of the other (regardless of the running positions in the field at which the two horses actually finish) at least 59% of the time (according to the computation that I indicated in my earlier post).

[davew]

When we say 59% we do not mean overall winners.

What we mean is that we will be correct 59% in our assesment that our selected horse will beat the other (for any position, regardelss if it won the race or not)


Is 59% your number or Overlays number? I am confused because you said you could pick 100% of the time whether horse A or B would be ahead of the other at the finish.

[ReplayRandall]

Interesting conversation.



Replay,

"Needing $5.80" does not actually solve the problem because, logically, as the price goes up (from average) the hit rate goes down.

I think I know where Delta L is going with this, but will hold back comment for now.


Do you still know where Delta is going with this? Maybe now is the time for you to help this thread along...........

[acorn54]

a horse that has no "holes" in it so to speak, to raise an element of doubt in the public's mind at large would probably not yield a positive expectation betting scenario.
there has to be some "confusion" among the bettors as to who to target in a race. this almost always happens in large race fields in my research.
also betting to place at ANY odds on my selections has shown very little positive results.

[TexasDolly]

This is not necessary for this thread, but since the winning frequencies of BPP came up a couple of times, this table is presenting them:


http://www.themindofagambler.com/bpp-winning-freq.jpeg



It seems to me that the # 1 beats the # 2 horse 60% + times now . The DL stats have the # 1 winning the race 31% to the @ 2 winning 20 % since we are not interested in the races they fail doesn't that satisfy the 59 % requirement ?
TD

[DeltaLover]

Here you can see the behaviour of the imaginary handicapper who is always correct when he selects one of the top two BPP horses:


http://themindofagambler.com/wf100p.jpeg



While here you can see the same metrics for more "down to earth" handicapper, who is correct 58% of the time:


http://themindofagambler.com/wf58p.jpeg



Note how the rating of first over second decreases...

[DeltaLover]

What you should immediately notice in this picture, is that the striking rate for the second player, is very close the top bpp alone!


31.12 vs 31.53

The 31.53 is coming from the previous table, showing the BPP rankings

[Robert Fischer]

So the key will be identifying when those 2 prime power horses are most likely to win with value.

In other words you will need to apply an additional filter(s) or factor(s) to reach your hypotheticals.

[JJMartin]

So the key will be identifying when those 2 prime power horses are most likely to win with value.

In other words you will need to apply an additional filter(s) or factor(s) to reach your hypotheticals.
When the pt odds appear to show value, the strike rate will be lower invariably. There is no magic number.

[TexasDolly]

I am missing something here. I thought it was only necessary to pick
the horse that beat the other horse 59 % of the time. The #1 appears to do that but of course he won't show a profit. DL.could you elaborate on the 2nd graph some ?
Thank you,
TD

[TrifectaMike]

A marriage of what you suggest with Dave's monte hall thingy might have possibilities.

Mike

[DeltaLover]

I am missing something here. I thought it was only necessary to pick
the horse that beat the other horse 59 % of the time. The #1 appears to do that but of course he won't show a profit. DL.could you elaborate on the 2nd graph some ?
Thank you,
TD

Excellent! What you say here, is the obvious...

The second graph was createde by a simulator who is assuming that in the case that one of the 1 - 2 horses is winning, we have a 58% chance to hit.

Nothing fancy as you can see..

Of course, it becomes obvious, that just having the highest chance of picking the best of the 1 - 2 is not enough. Always picking the 1 horse, fullfils this condition, but it fails miserably on ROI..

At this point we can see that we need another condition as well:

The rating of the selected As over the Bs should be as close as posible to the 1.53 rating of the perfect strategy.

So, we need to dicover a model that will maximize the correctness level AND have the ratio A / B as close to 1.53 as possible.

[TrifectaMike]

Dl,

Expected Win % for PP1 = Odds Ratio for PP1/ (Odds Ratio for PP1 + Odds Ratio for PP2)

Odds Ratios based on your table for long term stats and Expected Win % for PP1 is for a single event.

This will give you parameters for a uniform distribution from which you can sample to mimic Expected Win % for PP1.

Once you do this, you can look for another event to condition on or you can add PP3 to the mix and look at it as a Monty Hall problem with varying probabilities.

This is precious (Now, I know this thread will be locked).

Mike

[DeltaLover]

Dl,

Expected Win % for PP1 = Odds Ratio for PP1/ (Odds Ratio for PP1 + Odds Ratio for PP2)

Odds Ratios based on your table for long term stats and Expected Win % for PP1 is for a single event.

This will give you parameters for a uniform distribution from which you can sample to mimic Expected Win % for PP1.

Once you do this, you can look for another event to condition on or you can add PP3 to the mix and look at it as a Monty Hall problem with varying probabilities.

This is precious (Now, I know this thread will be locked).

Mike

I will look onto it tomorow ...

Precious :D

[TexasDolly]

Excellent! What you say here, is the obvious...

The second graph was createde by a simulator who is assuming that in the case that one of the 1 - 2 horses is winning, we have a 58% chance to hit.

Nothing fancy as you can see..

Of course, it becomes obvious, that just having the highest chance of picking the best of the 1 - 2 is not enough. Always picking the 1 horse, fullfils this condition, but it fails miserably on ROI..

At this point we can see that we need another condition as well:

The rating of the selected As over the Bs should be as close as posible to the 1.53 rating of the perfect strategy.

So, we need to dicover a model that will maximize the correctness level AND have the ratio A / B as close to 1.53 as possible.
Thanks Dl. So there was another condition besides the 58%. I now see the point of the graph.
TD

[Sapio]

Dl,

Expected Win % for PP1 = Odds Ratio for PP1/ (Odds Ratio for PP1 + Odds Ratio for PP2)

Odds Ratios based on your table for long term stats and Expected Win % for PP1 is for a single event.

This will give you parameters for a uniform distribution from which you can sample to mimic Expected Win % for PP1.

Once you do this, you can look for another event to condition on or you can add PP3 to the mix and look at it as a Monty Hall problem with varying probabilities.

This is precious (Now, I know this thread will be locked).

Mike

From Dl's table:

Odds Ratio for PP1 = 4695/(14890 - 4695) = .46052
Odds Ratio for PP2 = 3076/(14890 - 3076) = .26039

Expected Win % for PP1 = .46052/(.46052 + .26039) = .6388
Expected Win % for PP2= 1 - .6388 = .3612

So, in a matchup between PP1 and PP2, PP1 is expected to win 64% of the time and PP2 is expected to win 36% of the time.

Thomas Sapio

[Dave Schwartz]

Do you still know where Delta is going with this? Maybe now is the time for you to help this thread along...........

LOL - The first thing that came to mind was the quote that has been all over the web recently: "Not my circus..."

Actually, I am now having a difficult time following.

I thought he was moving towards some sort of Monty Hall problem solution. TM's responses look Bayesian to me as well. LOL

[Robert Fischer]

When the pt odds appear to show value, the strike rate will be lower invariably. There is no magic number.
There's a magic number, but it isn't bpp1+bpp2.

By structuring the question as a hypothetical where bpp1+2 meet a bunch of necessary conditions, what it really does is focus on the process of where to go from there.

The hard part is actually meeting those conditions in reality.

[HUSKER55]

hey guys and gals,

does anyone know where the numbers in tha table came from?

[DeltaLover]

hey guys and gals,

does anyone know where the numbers in tha table came from?


All the data are coming from BRIS

[JJMartin]

There's a magic number, but it isn't bpp1+bpp2.

By structuring the question as a hypothetical where bpp1+2 meet a bunch of necessary conditions, what it really does is focus on the process of where to go from there.

The hard part is actually meeting those conditions in reality.
I don't believe it is possible to improve upon this process to the level required. It is the same in principle as trying to determine which of the top 2 pt fav horses will win by using some criteria unknown to everyone else. Either one of the top 2 choices alone always produces a negative roi, aside from a utilizing a crystal ball you will not change this reality with some arbitrary adjustment. Imo, this approach will never work because it is based on a faulty premise.

[Sapio]

I don't believe it is possible to improve upon this process to the level required. It is the same in principle as trying to determine which of the top 2 pt fav horses will win by using some criteria unknown to everyone else. Either one of the top 2 choices alone always produces a negative roi, aside from a utilizing a crystal ball you will not change this reality with some arbitrary adjustment. Imo, this approach will never work because it is based on a faulty premise.

Hi JJmartin,

You might correct in absolute terms, but this is not a game of absolutes. It is a game based on "relative" correctness... small incremental steps in a good direction.

Exploring a process, even if it ends in failure can produce some useful results... better insights and better methods.

For example, given what DL has purposed and TM's posts has me thinking of attepting to determine a relative overlay between PP1 and PP2.

This is a worthwhile exercise.

Thomas Sapio

[JJMartin]

Hi JJmartin,

Exploring a process, even if it ends in failure can produce some useful results... better insights and better methods.

Thomas Sapio

Agreed.

[whodoyoulike]

This is not necessary for this thread, but since the winning frequencies of BPP came up a couple of times, this table is presenting them:


http://www.themindofagambler.com/bpp-winning-freq.jpeg


Wouldn't the %'s be distorted because you're lumping different field sizes together? I would think the winning %'s would be different if you were to stratify the database by field size (i.e., 7, 8 and 9 horse fields which are common field sizes [not necessary to stratify all but just these three for your hypothesis]). If the results were similar, it would provide additional support for what you're attempting to demonstrate.

Am I interpreting your table correctly that for field sizes of 9 and above the #1 BPP win less than 3.5%? Which kind of provides support to my thinking that full field sizes encourages competitive racing where the "favorites" don't necessarily win and the winners have nice payoffs. And, I realize that the #1 BPP is not necessarily the race favorite but, it frequently is.

Anyone else interpret the results of the > 8 horse fields differently?

[TexasDolly]

Whodo, I think that means it is the 8th rank of the BPP
but that there are fewer races with a #8 . As I understood it the #1 ranking is included for 8 horse fields in the first row as well all the other size fields. Should they be separated, probably.
TD

[whodoyoulike]

Thanks, you're correct. I think I understood it on my first read. My second paragraph was an after thought which came to me as I was reviewing the table just before I hit submit. Thanks again for the clarification.

[HUSKER55]

Boy am I having a bad week or what. I thought the first colum was the rank of the Prime Power to the field. and not the size of the field.

[Overlay]

Boy am I having a bad week or what. I thought the first colum was the rank of the Prime Power to the field. and not the size of the field.
The first column is the Prime Power rank. The way that I read the table, the entire sample consisted of 14,890 races. Every one of those 14,890 races had a minimum of three horses. 14,864 of the 14,890 races had four horses running. 14,661 of the 14,890 had five. 13,821 had six. 11,876 had seven. 9,148 had eight. 6,322 had nine. 4,106 had ten. 1,967 had eleven. 991 had twelve. 84 had 13. 40 had 14.

The horse that ranked first (regardless of field size) won 4,695 of its 14,890 races, for a winning percentage of 31.53%. The horse that ranked second (regardless of field size) won 3,076 of its 14,890 races, or 20.66%. The horse that ranked third (regardless of field size) won 2,113 of its 14,890 races, or 14.19%. The horse that ranked fourth (in the 14,864 fields that had four or more starters) won 1,606 of those 14,864 races, or 10.80%. Similarly, the horse that ranked fifth (in the 14,661 fields that had five or more starters) won 1,210 of those 14,661 races, or 8.25% . The horse that ranked sixth (in the 13,821 races in which there were six or more starters) won 856 of those 13,821 races, or 6.19%. And so forth down to the line for fourteen-horse fields, where the horse that ranked fourteenth won none of the 40 races in which it appeared (0.0%).

[DeltaLover]

The first column is the Prime Power rank. The way that I read the table, the entire sample consisted of 14,890 races. Every one of those 14,890 races had a minimum of three horses. 14,864 of the 14,890 races had four horses running. 14,661 of the 14,890 had five. 13,821 had six. 11,876 had seven. 9,148 had eight. 6,322 had nine. 4,106 had ten. 1,967 had eleven. 991 had twelve. 84 had 13. 40 had 14.

The horse that ranked first (regardless of field size) won 4,695 of its 14,890 races, for a winning percentage of 31.53%. The horse that ranked second (regardless of field size) won 3,076 of its 14,890 races, or 20.66%. The horse that ranked third (regardless of field size) won 2,113 of its 14,890 races, or 14.19%. The horse that ranked fourth (in the 14,864 fields that had four or more starters) won 1,606 of those 14,864 races, or 10.80%. Similarly, the horse that ranked fifth (in the 14,661 fields that had five or more starters) won 1,210 of those 14,661 races, or 8.25% . The horse that ranked sixth (in the 13,821 races in which there were six or more starters) won 856 of those 13,821 races, or 6.19%. And so forth down to the line for fourteen-horse fields, where the horse that ranked fourteenth won none of the 40 races in which it appeared (0.0%).

exactly :ThmbUp:

[ReplayRandall]

The first column is the Prime Power rank. The way that I read the table, the entire sample consisted of 14,890 races. Every one of those races had a minimum of three horses. 14,864 of the 14,890 had four horses running. 14,661 of them had five. 13,821 had six. 11,876 had seven. 9,148 had eight. 6,322 had nine. 4,106 had ten. 1,967 had eleven. 991 had twelve. 84 had 13. 40 had 14.

The horse that ranked first (regardless of field size) won 4,695 of its 14,890 races. The horse that ranked second (regardless of field size) won 3,076 of its 14,890 races. The horse that ranked third (regardless of field size) won 2,113 of its 14,890 races. The horse that ranked fourth (in the 14,864 fields that had four or more starters) won 1,606 of those 14,864 races. Similarly, the horse that ranked fifth (in the 14,661 fields that had five or more starters) won 1,210 of those 14,661 races. The horse that ranked sixth (in the 13,821 races in which there were six or more starters) won 856 of those 13,821 races. And so forth down to the line for fourteen-horse fields, where the horse ranked fourteenth won none of the 40 races in which it appeared.

The chart is not a true representation of what each Prime power rank win rate is, against a specific number of horses in a field. For instance, we are handicapping a race of 12 starters, so by looking at the chart, we take out the 40 starts of 14 horse fields and 84 starts of 13 horse fields. Now we take 14,890-(40+84)=14,766, and that is now the current sample for all races of at least 12 horse fields. However, we do not know the number of wins PP1 had when we threw out those 120 races, thus skewing the rest of the table. The win rate is not reliable because it is not specific enough for EXACT field sizes. Other posters have mentioned this dilemma and I am convinced they are correct...........more filtering is needed.

[whodoyoulike]

So, before I get all screwed up again, reading the table and attempting to get some value from the info, one wouldn't want to wager to win the #1 BPP in a 14 horse field.

Does anyone disagree?


Delta Lover, is it still possible to stratify the results specifically only for exactly 7 - 14 horse fields in each query? I'm still curious of the results if it's too difficult that's okay. I'll understand.

[DeltaLover]

The chart is not a true representation of what each Prime power rank win rate is, against a specific number of horses in a field. For instance, we are handicapping a race of 12 starters, so by looking at the chart, we take out the 40 starts of 14 horse fields and 84 starts of 13 horse fields. Now we take 14,890-(40+84)=14,766, and that is now the current sample for all races of at least 12 horse fields. However, we do not know the number of wins PP1 had when we threw out those 120 races, thus skewing the rest of the table. The win rate is not reliable because it is not specific enough for EXACT field sizes. Other posters have mentioned this dilemma and I am convinced they are correct...........more filtering is needed.

What you say here is correct and we have talk about it in the past. The important conclusion we can derive from this picture is that BPP is a very reliable metric, which represents the ability of each horse in the race to a very good extend... I have test BPP specificaly for fields of 5 - 6 - 7 etc horses individually and in all cases it behaves similarly to the mixed number of starters case.

For the purpose of this thread, we do not care about the precise winning percentages (since we do not use them anywhere) but for a metric that we have good reason to believe that it is reliable for what it measures.

[Overlay]

So, before I get all screwed up again, reading the table and attempting to get some value from the info, one wouldn't want to wager to win the #1 BPP in a 14 horse field.

Does anyone disagree?
No, the table is saying that you wouldn't want to wager on the horse that ranked last (14th out of 14) in BPP in a 14-horse field, since that horse failed to win any of the 40 races in which it appeared.

The horse that ranked first (without taking field size into consideration) won 4,695 races of the 14,890 in which it appeared . However, that does not tell you how many of those 4,695 victories were against three-horse fields, four-horse fields, five-horse fields, etc. One would expect that the overall winning percentage of 31.53% for the top-ranked horse would have been higher than that against smaller fields, and would have decreased as field size grew larger, but you can't determine what that winning percentage against each differing field size would have been based only on the data given in the table. (The same situation exists with the victories recorded by the second-ranked horse, the third-ranked horse, etc.) (I believe that this was the point that ReplayRandall was making, also.)

[Vinman]

The BRIS based spreadsheet methodology that I use measures the impact value of the top ranked Prime Power horse. This is typically between 2.0 and 3.0 depending on whether we're dealing with dirt sprints or routes, or turf sprints or routes. I track each separately. I also track the I.V. of any horse(s) that scores within one point of the top ranked Prime Power horse in that race, if any, those that are two points off the top horse, three points off, etc. all the way down through the entire field. While I'm all in favor of focusing on the top two Prime Power horses, that doesn't really go far enough in my opinion. You obviously have races where the top two horses are at say, 121.9 and 120.7 versus a 130.4 over 123.6 scenario. So rather than simply measure the effectiveness of "the top two Prime Power horses", I believe it is even more important to focus on how many POINTS a horse ranks behind the "Alpha horse" in Prime Power. It makes a big difference.

Vinman

[whodoyoulike]

No, the table is saying that you wouldn't want to wager on the horse that ranked last (14th out of 14) in BPP in a 14-horse field, since that horse failed to win any of the 40 races in which it appeared...



Thanks.

[DeltaLover]

So, before I get all screwed up again, reading the table and attempting to get some value from the info, one wouldn't want to wager to win the #1 BPP in a 14 horse field.

Does anyone disagree?


Delta Lover, is it still possible to stratify the results specifically only for exactly 7 - 14 horse fields in each query? I'm still curious of the results if it's too difficult that's okay. I'll understand.

In this picture you can see how the top BPP behaves according to the field size:


http://themindofagambler.com/top-bpp.jpeg


Of course this data are not directly related to the topic of this thread!

I can see that this type of statistics create some interest, although here we do not try to discover betting approaches based in the BPP alone, instead we use BPP as a good enough indicator that can serve well the assumption of limiting our selections to two specific horses.

The effectiveness of the specific metric (BPP in this case) is clearly secondary for what we try to do here, which is developing a methodology to resemble the perfect "imaginary" handicapper who is able to always select the correct horse, as much as possible..

[HUSKER55]

thank you guys and gals!

[ReplayRandall]

In this picture you can see how the top BPP behaves according to the field size:


http://themindofagambler.com/top-bpp.jpeg


Of course this data are not directly related to the topic of this thread!

I can see that this type of statistics create some interest, although here we do not try to discover betting approaches based in the BPP alone, instead we use BPP as a good enough indicator that can serve well the assumption of limiting our selections to two specific horses.

The effectiveness of the specific metric (BPP in this case) is clearly secondary for what we try to do here, which is developing a methodology to resemble the perfect "imaginary" handicapper who is able to always select the correct horse, as much as possible..

Thanks DL for indulging me, please continue........

[DeltaLover]

So rather than simply measure the effectiveness of "the top two Prime Power horses", I believe it is even more important to focus on how many POINTS a horse ranks behind the "Alpha horse" in Prime Power. It makes a big difference.


Do you have some data to support your case? What exactly do you mean by 'makes a big difference'? Also note, that here I do not try to measure the effectiveness of "the top two Prime Power horses", what I am trying to is very concrete: Discover the characteristics of a model, that will mimic as much as possible the perfect handicapper as I described it in the begining of this thread. Things like the point speads, the size of the field and any other factor (including pace, recency, workouts etc) can be used in the creation of the model, but are not needed for the description of the characterists of the model, which as I said above are (1) Maximize the times when we are correct about A vs B and (2) Keep the rating A/B near 1.53

[lansdale]

Hi DL,

This is an intriguing problem. Purely as speculation, I was wondering about the utility of Shannon's outguessing game, if it's possible this problem has a game theory solution. The great scientist designed a very primitive computer to play the old game of matching pennies - and the machine invariably won, beating some of the leading scientists in the world.

The math people here can tell my instantly whether this is wrong - but, since I'm assuming the distribution is uniform, I'm wondering if it would be possible to mimic the distribution, but do so randomly or at least in a manner that approaches randomness - as Shannon proved humans are incapable of doing.

I have a more obvious suggestion, but I'll wait for some feedback on the viability of this first.

Cheers,

lansdale

[DeltaLover]

Hi DL,

This is an intriguing problem. Purely as speculation, I was wondering about the utility of Shannon's outguessing game, if it's possible this problem has a game theory solution. The great scientist designed a very primitive computer to play the old game of matching pennies - and the machine invariably won, beating some of the leading scientists in the world.

The math people here can tell my instantly whether this is wrong - but, since I'm assuming the distribution is uniform, I'm wondering if it would be possible to mimic the distribution, but do so randomly or at least in a manner that approaches randomness - as Shannon proved humans are incapable of doing.

I have a more obvious suggestion, but I'll wait for some feedback on the viability of this first.

Cheers,

lansdale


I do not think that GT is applicable to the particular problem. This is not to say that GT cannot be used in horse racing, but I would limit it to the intentions of the connections and do not try to apply it to pure handicapping. The way I defined the problem, does not introduce any rational decision makers where GT might be applicable, so I really cannot see any possible application of it. It will be helpful though, if you explain in more detail the extensive and normal forms of your proposed GT based model.

[whodoyoulike]

Thanks for providing the latest table. It provided me info on field sizes, competitiveness and performance of favorites.

Regarding your thread premise, are you suggesting the odds requirements has to be XX if you are using #1 or #2 BPP's? It seems this a handicapping angle but, you're using the angle prior to handicapping the race (and, I'm uncertain you will eventually even handicap the race).

[DeltaLover]

Regarding your thread premise, are you suggesting the odds requirements has to be XX if you are using #1 or #2 BPP's?

Not at all

What I am saying is that the model needs to always select a horse between A and B while maitaining a selection rate of 1.5, meaning of each 3 A horses it should pick 2 Bs.

Odds are not considered at all in this system

[davew]

I was assiming that the charts presented was from a database kept by DeltaLover. If this is not the case, and from an expanded report from Bris (where can I get this report), does anyone know what they do when there is not a bpp on a horse (first time starter in NA)?

My personal opinion is that the distribution of the scores in the field is more important than which is ranked 1,2,3.... an example of a couple 5 horse races

race 1- 131.2, 126.4, 117.3, 117.1, 103.2
race 2- 124.4, 124.1, 123.9, 123.5, 117.4

and what if the odds on these 5 were
3/5, 5/2, 10, 5, 6

would you still want to limit bets to top 2 bpp?

What I would find very interesting is match-ups between 2 horses and the percentage time the highest finished ahead of the other when they were higher by exactly 0.1, 0.2, 0.3,..... 1.1, 1.2,..... 5.1, 5.2, .... 15.1, 15.2, ...
and the frequency this occurred.

[DeltaLover]

I was assiming that the charts presented was from a database kept by DeltaLover. If this is not the case, and from an expanded report from Bris (where can I get this report), does anyone know what they do when there is not a bpp on a horse (first time starter in NA)?




All the data I am using are from my own databases...

[DeltaLover]

would you still want to limit bets to top 2 bpp?


I am just trying to discover a model that will resemble the perfect handicapper as I have described him above. This methodology can be useflull for more adnvanced handicapping once you perfect it.

[traynor]

I am just trying to discover a model that will resemble the perfect handicapper as I have described him above. This methodology can be useflull for more adnvanced handicapping once you perfect it.

Indeed.

[Poindexter]

DL as a second filter how about a performance rating. In another thread(a while back) it was discussed that you can use convert the cr to a numerical rating. A cr of 117 would be a 90, add 2 points for every point above and subtract 2 points for every point below. So if they have a cr of 115 that equals an 86. If it is a cr of 119 that equals 94. So take this performance rating average it with the speed rating and I feel you get a more accurate figure than a speed figure alone. EG a cr of 115(which converts to an 86) with a speed rating of 92 would equate to a performance rating of 89.

So if you want a way of separating the top 2 Bris power rating horses, compare the last 2 performance ratings for these horses. If horse A has both performance ratings better than horse B he would be a double advantage horse. If horse A has a better last performance rating but a weaker prior race performance rating he would be a single advantange horse. I would then look at the roi of
Double advantage
Single advantage
Single disadvantage
Double disadvantage

The objective is to determine which of the two horses offers a better bet, the one that is theoretically faster or the one that is theoretically slower or is the roi roughly the same no matter what. I assume you have the last 2 crs and speed ratings in your database or else you will not be able to do this.

[barn32]

I've had too much turkey and I'm sleepy. Two simple questions:

Why do the percentages in this table add up to 113.03?

Why is the 12th ranked horse higher than the 11th, 10th
9th and 8th ranked horses?

http://oi57.tinypic.com/55nnlx.jpg

[DeltaLover]

I've had too much turkey and I'm sleepy. Two simple questions:

Why do the percentages in this table add up to 113.03?

Why is the 12th ranked horse higher than the 11th, 10th
9th and 8th ranked horses?

http://oi57.tinypic.com/55nnlx.jpg

These percentage do not have to add up to 100!

An simple example so you can understand it:

Lets assume that we have 100 races, 98 of them consist of exactly two horses, while the remaining two race of three. Out of the 98 two horse fields we have 49 A and 49 B while both of the other races are won by the third horse.

In this case our percentages will look like this:

Rank 1 -> 49 out of 98 = 50%
Rank 2 -> 49 out of 98 = 50%
Rank 3 -> 2 out of 2 = 100%

Adding the percentages up gives us 200%

As you can see, it does not make any sense to add the percentages up, since the number of the horses is different in each race.

[DeltaLover]

Why is the 12th ranked horse higher than the 11th, 10th
9th and 8th ranked horses?



Obviosuly the insufficient sample size is to be blamed for this anomaly. If we add more races containing 12 and more horses, most likely this spot will regress to fit the behaviour of the rest of the curve.

[TrifectaMike]

Too muich confusion....

Put all the data, table, criteria, aside for the moment

Let 2 represent that PP2 beats PP1
Let 1 represent that PP1 beats PP2

In a series of races, we have the following string

1 2 1 2 1 1 1 2 2 1 1 1 1 2 1 2 1 2 1 1 ....

What DL is attempting to do is to somehow mimic this string (his imaginary perfect selector) with a not so perfect selector, but good enough to be within a range of goodness.

Precious Mike

[DeltaLover]

Too muich confusion....

Put all the data, table, criteria, aside for the moment

Let 2 represent that PP2 beats PP1
Let 1 represent that PP1 beats PP2

In a series of races, we have the following string

1 2 1 2 1 1 1 2 2 1 1 1 1 2 1 2 1 2 1 1 ....

What DL is attempting to do is to somehow mimic this string (his imaginary perfect selector) with a not so perfect selector, but good enough to be within a range of goodness.

Precious Mike

Exactly Mr. Precious!

[JJMartin]

Too muich confusion....

Put all the data, table, criteria, aside for the moment

Let 2 represent that PP2 beats PP1
Let 1 represent that PP1 beats PP2

In a series of races, we have the following string

1 2 1 2 1 1 1 2 2 1 1 1 1 2 1 2 1 2 1 1 ....

What DL is attempting to do is to somehow mimic this string (his imaginary perfect selector) with a not so perfect selector, but good enough to be within a range of goodness.

Precious Mike
The way I see it, he is trying to add "something" to pp2 to make it #1 and when the "something" does not exist, you bet pp1 by default. The problem is that you are tampering with a rating that is essentially a composite factor that was tested and refined probably over 100,000 races, maybe even a million. The bpp is already the best it can be (in theory), any thing else you would do to it will probably undermine its integrity and lower its accuracy. Not trying to be discouraging or critical just making an observation.

When pp2 beats pp1 in a race, it could be because pp1 suffered an injury in training or stumbled at the start or any variation of circumstances impossible to determine or predict.

[hardboot]

This is not necessary for this thread, but since the winning frequencies of BPP came up a couple of times, this table is presenting them:


http://www.themindofagambler.com/bpp-winning-freq.jpeg


Thanks for the chart.... slightly off the main topic, but do you happen to have any possibly more specific
ie ---- claiming, alw (split even more as nw2 etc), distance, surface.

I bet only claiming sprints, for example, as I've found my predictive ability to be better with limits.

thanks

[DeltaLover]

Thanks for the chart.... slightly off the main topic, but do you happen to have any possibly more specific
ie ---- claiming, alw (split even more as nw2 etc), distance, surface.


We can easily filter these reports by any additional conditions (like classification, distance or surface).. Still, this approach belongs to lower levels of handicapping that are not part of the macro-handicapping I am performing here. At this point, I am trying to keep things as simple as possible, in other words: KISS

[barn32]

"What DL is attempting to do is to somehow mimic this string
(his imaginary perfect selector) with a not so perfect selector,
but good enough to be within a range of goodness.

Precious Mike"

Is this even possible? If it is possible, then what's the first step?

"We can easily filter these reports by any additional
conditions (like classification, distance or surface)..
Still, this approach belongs to lower levels of
handicapping that are not part of the
macro-handicapping I am performing here. At this
point, I am trying to keep things as simple as possible,
in other words: KISS

Delta Lover"

So, RE the above from Mike: are you onto something, or are
you still trying to find (or get help finding) an answer?

[DeltaLover]

"What DL is attempting to do is to somehow mimic this string
(his imaginary perfect selector) with a not so perfect selector,
but good enough to be within a range of goodness.

Precious Mike"

Is this even possible? If it is possible, then what's the first step?

"We can easily filter these reports by any additional
conditions (like classification, distance or surface)..
Still, this approach belongs to lower levels of
handicapping that are not part of the
macro-handicapping I am performing here. At this
point, I am trying to keep things as simple as possible,
in other words: KISS

Delta Lover"

So, RE the above from Mike: are you onto something, or are
you still trying to find (or get help finding) an answer?

Hmm.. Let's say that I need some help... Any ideas?

[thaskalos]

Hmm.. Let's say that I need some help... Any ideas?

I hope this helps:

http://www.amazon.com/Complete-Psychology-mathematics-Tactics-Winning/dp/B001PXTRRI/ref=sr_1_2?ie=UTF8&qid=1417196060&sr=8-2&keywords=play+poker+quit+work+and+sleep+till+noon

[DeltaLover]

I hope this helps:

http://www.amazon.com/Complete-Psychology-mathematics-Tactics-Winning/dp/B001PXTRRI/ref=sr_1_2?ie=UTF8&qid=1417196060&sr=8-2&keywords=play+poker+quit+work+and+sleep+till+noon

One of the very first books I have ever read about poker... Dated since it is about draw but still has interesting information, especially about tell detection..

Thanks anyway mr Thask :ThmbUp:

[thaskalos]

One of the very first books I have ever read about poker... Dated since it is about draw but still has interesting information, especially about tell detection..

Thanks anyway mr Thask :ThmbUp:

Wow...we DO have something in common. :)

[DeltaLover]

Wow...we DO have something in common. :)

Τα μεγαλα πνευματα συναντωνται :)

[Red Knave]

The problem is that you are tampering with a rating that is essentially a composite factor that was tested and refined
Right. It is a very high level composite that includes, I suspect, a great many well weighted handicapping factors. So we probably need to find some non-handicapping factor(s) or maybe some factor that points to losing rather than winning to complement the BPP rankings.

When pp2 beats pp1 in a race, it could be because pp1 suffered an injury in training or stumbled at the start or any variation of circumstances impossible to determine or predict.I suspect that these events would even out over a large enough sample of races, i.e. it would happen to both PP1 and PP2 an equal number of times. However, we don't know any of this before the race starts so it can't help us solve the problem at hand. We need something we can use before the race starts.

pompous Red :)

[ReplayRandall]

Hmm.. Let's say that I need some help... Any ideas?


Maybe implementing a deadline for this thread to come to a conclusion. I'll give you until Sun. Nov.30th- 9PM-EST. I believe this will help you immensely with focus, rising above your comfort level, where the biggest gains or breakthroughs materialize.......

[barn32]

I hope this helps:
Play Poker, Quit Work,
and Sleep Til Noon!

I have a pristine hardcover copy of this book that
I got for $6.00 when one of the gambling bookstores went
out of business back in the 90s.

There were ten copies available, and I'm kicking myself to
this day for not buying all ten of them.

[davew]

I hope this helps:

http://www.amazon.com/Complete-Psychology-mathematics-Tactics-Winning/dp/B001PXTRRI/ref=sr_1_2?ie=UTF8&qid=1417196060&sr=8-2&keywords=play+poker+quit+work+and+sleep+till+noon


haven't read that one, but this one mught give you some ideas

http://www.amazon.com/Calculated-Bets-Computers-Gambling-Mathematical/dp/0521009626/ref=sr_1_1?ie=UTF8&qid=1417197870&sr=8-1&keywords=calculated+bets

[DeltaLover]

Maybe implementing a deadline for this thread to come to a conclusion. I'll give you until Sun. Nov.30th- 9PM-EST. I believe this will help you immensely with focus, rising above your comfort level, where the biggest gains or breakthroughs materialize.......

I see it the dead line to be a little aggressive.. The weekend has aleady started and is full of racing, handicapping, replay watch and betting leave little time even to eat and sleep, setting aside completing the this thread... We have plently of time ahead of us during the dark days of the upcoming weeks :)

[Greyfox]

Τα μεγαλα πνευματα συναντωνται :)

The Great Spirit Meet? :confused:

[DeltaLover]

The Great Spirit Meet? :confused:

So, you can read Greek? I am impressed! :)

[thaskalos]

haven't read that one, but this one mught give you some ideas

http://www.amazon.com/Calculated-Bets-Computers-Gambling-Mathematical/dp/0521009626/ref=sr_1_1?ie=UTF8&qid=1417197870&sr=8-1&keywords=calculated+bets
I purchased this book on March 23...2010.

http://www.amazon.com/Calculated-Bets-Computers-Gambling-Mathematical/dp/0521009626/ref=sr_1_1?ie=UTF8&qid=1417197870&sr=8-1&keywords=calculated+bets

When I say that I've read every single worthwhile gambling book there is...I really mean it. :)

[lansdale]

So, you can read Greek? I am impressed! :)

Great minds think alike.

[thaskalos]

Yeah...it means "great minds meet". But DeltaLover has always overestimated my mind. :)

[lansdale]

I have to say I'm in agreement with those here don't see the solution to this macro problem as a micro answer (more information, fundamentals) of the kind that you seem to be soliciting. I think it has to have a macro answer. But I'm now more convinced that there is no solution. The PP win distribution seems to me close to a power law distribution, and as such, can't be overcome, any more easily than the laws of gravity.

You didn't really respond to the Shannon game theory idea in a granular way, so I'll elaborate, although, as I said, it's speculative. You must be familiar with the game of Prisoner's Dilemma. Two players have a choice to either agree or disagree, defect or cooperate, but don't know the other's choice in advance. In Shannon's game of man vs. machine, the winning play was to always defect. The 'machine', a very primitive, limited 'computer' beat top scientists decisively. Why? Because their decisions reflected patterns, non-randomness, in a word cardinality - that which can be exploited by an opponent, while itself, as a machine was able to maintain perfectly random decision-making. This is fundamentally identical to the way in which high-level poker players are now using Bayesian, and game theory techniques to defeat unwitting opponents.

So my counterfactual is this - what if we think of PP1 and PP2 as an 'unfair' 60/40 coin toss. We play that exact distribution uniformly, but with a simmed series of randomized decisions. On reflection, I would have to admit that this will not enhance our ability to mimic the ideal handicapper, but it should be a test for evidence of cardinality, which we should be able to exploit, in the public line. My guess though, is that, although the 'powerlaw-ness' of Bris Prime Power is likely augmented by the globalization of the sample, it might be possible to uncover cardinality at specific tracks.

[traynor]

So, you can read Greek? I am impressed! :)

As an alternative, copy and paste it into Google:
Translate Τα μεγαλα πνευματα συναντωνται

[lansdale]

I have to say I'm in agreement with those here don't see the solution to this macro problem as a micro answer (more information, fundamentals) of the kind that you seem to be soliciting. I think it has to have a macro answer. But I'm now more convinced that there is no solution. The PP win distribution seems to me close to a power law distribution, and as such, can't be overcome, any more easily than the laws of gravity.

You didn't really respond to the Shannon game theory idea in a granular way, so I'll elaborate, although, as I said, it's speculative. You must be familiar with the game of Prisoner's Dilemma. Two players have a choice to either agree or disagree, defect or cooperate, but don't know the other's choice in advance. In Shannon's game of man vs. machine, the winning play was to always defect. The 'machine', a very primitive, limited 'computer' beat top scientists decisively. Why? Because their decisions reflected patterns, non-randomness, in a word cardinality - that which can be exploited by an opponent, while itself, as a machine was able to maintain perfectly random decision-making. This is fundamentally identical to the way in which high-level poker players are now using Bayesian, and game theory techniques to defeat unwitting opponents.

So my counterfactual is this - what if we think of PP1 and PP2 as an 'unfair' 60/40 coin toss. We play that exact distribution uniformly, but with a simmed series of randomized decisions. On reflection, I would have to admit that this will not enhance our ability to mimic the ideal handicapper, but it should be a test for evidence of cardinality, which we should be able to exploit, in the public line. My guess though, is that, although the 'powerlaw-ness' of Bris Prime Power is likely augmented by the globalization of the sample, it might be possible to uncover cardinality at specific tracks.

Ordinality vs. Cardinality

[DeltaLover]

Ordinality vs. Cardinality

Can you post a concrete example?

[whodoyoulike]

Hmm.. Let's say that I need some help... Any ideas?

I've looked at your table over the last day or so and I think it's just an interesting tidbit of info. Similar to the favorites have won approx. 33% since I've been handicapping. Neither will help you in your handicapping horses. The reason I'm saying this, is because both bits of info don't consider the conditions of the current race which you're currently handicapping.

They're basically meaningless numbers.

Now you're database can provide some useful info as others have suggested regarding the stratification of variables. It won't help you with a particular race but should provide additional understanding of the dynamics of handicapping a race.

[DeltaLover]

Neither will help you in your handicapping horses.

I disagree...

I like to talk with pure facts and not opinions. If you cannot justify your 'opinions' with hard data, whatever you can claim is just that, an opinion...

[whodoyoulike]

I disagree...

I like to talk with pure facts and not opinions. If you cannot justify your 'opinions' with hard data, whatever you can claim is just that, an opinion...

You omitted the rest of the paragraph.

... The reason I'm saying this, is because both bits of info don't consider the conditions of the current race which you're currently handicapping.

They're basically meaningless numbers...



If you're able to show how your numbers consider the current conditions of the current race. I'll consider that it just may be a fact. Until then, it still appears a meaningless number. And, you are correct this is my opinion.

[lansdale]

Can you post a concrete example?

Do you really need an example? I think the idea that you profit from having information about your opponent's future decision(s) without the converse being true is self-evident. High-level pokers player who use a couple of heuristics to identify opponents' playing styles, while randomizing their own, have an edge. And I think you understand, better than most, that individual hands, like individual races, are irrelevant. As you said in one thread, in racing, the horses are just randomizers, the point is to play optimally. The same is true of poker - the cards are just randomizers. As I've posted before, the University of Alberta, which has studied all the Holdem variants exhaustively, has found the 'long run' of NLHE, the time required to determine the quality of a given poker player, is 100k hands - ca. two years of full-time play. Individual hands are meaningless, and the idea of a 'showdown' is a delusion.

But, if you follow what I was saying, the Shannon analogy is something that needs to be tested, unless the math guys agree that it has no possible validity or relevance either to the thread or to the potential utility of Prime Power as a handicapping tool.

[traynor]

I disagree...

I like to talk with pure facts and not opinions. If you cannot justify your 'opinions' with hard data, whatever you can claim is just that, an opinion...

I had hoped you were going to post some hard data indicating process whatever of meta handicapping that would enable one to correctly distinguish which of the top two PP selections is more likely to prevail in a given race (or a large number of given races).

It seems you are asking for strategies, yet reluctant to explicitly state such a goal. Do you have a solution, or are you asking if anyone else does (and is willing to toss it on the table)?

[DeltaLover]

I had hoped you were going to post some hard data indicating process whatever of meta handicapping that would enable one to correctly distinguish which of the top two PPP selections was more likely to prevail in a given race.

It seems you are asking for strategies, yet reluctant to explicitly state such a goal. Do you have a solution, or are you asking if anyone else does (and is willing to toss it on the table)?

You missed the point of your thread.. Try again

[DeltaLover]

Do you really need an example?


Yes, a concrete one.. No theories please..

[lansdale]

Yes, a concrete one.. No theories please..

I thought the poker example was pretty concrete. The Shannon machine I cited is also a concrete example of using an opponent's behavioral pattern to predict their response. Since this example does demonstrate my point, I don't know that's there's much more to say.

BTW, I'm not claiming this is viable. As I said, the analogy is purely speculative. Maybe TM or Thomas Sapio can take a look.

[DeltaLover]

I thought the poker example was pretty concrete. The Shannon machine I cited is also a concrete example of using an opponent's behavioral pattern to predict their response. Since this example does demonstrate my point, I don't know that's there's much more to say.

BTW, I'm not claiming this is viable. As I said, the analogy is purely speculative. Maybe TM or Thomas Sapio can take a look.

As hard as I have thought about in the past, I was never able to understand how I can apply GT to the problem we are discussing here. The poker example is obvious as there are many other similars that invlolve multiple parties who try to make a decission; still I do not think that we have similar case here..

If you have a concrete application of GT to handicapping, I would like to hear about...

[traynor]

You missed the point of your thread.. Try again

No thanks. BTW, it's not my thread, it's yours.

[DeltaLover]

No thanks. BTW, it's not my thread, it's yours.
ok, np

[TexasDolly]

Hi DL,
Your post # 45 has some numbers which puzzle me. The number of wins for PP #2 is (in the roi=.99 ex.) listed as 1974 and for PP #! is 2711. At a 20 % win rate this would require betting 1974/.2 = 9870 races. For the PP #1 2711/.3 = 9037 races.
Assuming my observation is correct, how can we accommodate that no. of races betting only one in each race ?
Thank you,
TD

[DeltaLover]

Hi DL,
Your post # 45 has some numbers which puzzle me. The number of wins for PP #2 is (in the roi=.99 ex.) listed as 1974 and for PP #! is 2711. At a 20 % win rate this would require betting 1974/.2 = 9870 races. For the PP #1 2711/.3 = 9037 races.
Assuming my observation is correct, how can we accommodate that no. of races betting only one in each race ?
Thank you,
TD

Where did you find the 20% win rate? This was never mentioned here. What I list, is the winners for a 58% correctness, that has nothing to do with winning %

[TexasDolly]

Where did you find the 20% win rate? This was never mentioned here. What I list, is the winners for a 58% correctness, that has nothing to do with winning %

The win rates for the top 2 BPP's were taken from your earlier
post # 35 where each BPP rank win % was listed. I assumed those held somewhat even with fewer plays. Otherwise, it seems to me to arrive at those number of wins the win % would need to be higher than the numbers you posted in post # 35.
Thank you,
TD

[barn32]

As I've posted before, the University of Alberta, which has studied all the Holdem variants exhaustively, has found the 'long run' of NLHE, the time required to determine the quality of a given poker player, is 100k hands - ca. two years of full-time play. Poki-bot? :cool:

[Red Knave]

Your post # 45 has some numbers which puzzle me. The number of wins for PP #2 is (in the roi=.99 ex.) listed as 1974 and for PP #1 is 2711
These numbers are assuming that you, the handicapper, were able to correctly get the correct order of the BPP top 2 58% of the time.

There are some small anomalies in the numbers which affect the results in a small way but not enough to derail the idea here.
For example, 58% of 4695 is not 2711, it is 2723. The total number of races (or races with a winner that has a BPP number) in the 100% correct list is 4695 + 3076 + 7103 = 14874.
In the 58% correct list it is 2711 + 1974 + 10369 = 15054. A difference of 180 which sounds suspiciously like a transposition error somewhere.
In any event, getting the order of PP1 and PP2 correct less than 60% of the time still allows for a .99 ROI which I find very interesting.

[TexasDolly]

These numbers are assuming that you, the handicapper, were able to correctly get the correct order of the BPP top 2 58% of the time.

There are some small anomalies in the numbers which affect the results in a small way but not enough to derail the idea here.
For example, 58% of 4695 is not 2711, it is 2723. The total number of races (or races with a winner that has a BPP number) in the 100% correct list is 4695 + 3076 + 7103 = 14874.
In the 58% correct list it is 2711 + 1974 + 10369 = 15054. A difference of 180 which sounds suspiciously like a transposition error somewhere.
In any event, getting the order of PP1 and PP2 correct less than 60% of the time still allows for a .99 ROI which I find very interesting.

Hi Red Knave,
To be honest I didn't follow your post very well. My post was intended to point out that in order to have the number of winners in post # 45 (with the win rates shown in post # 35)
there would have needed to be almost 19000 races. Or, in the alternative, the win % would have needed to increase a good bit. Since the number of races shown was only 15000+ I was curious as to what we were seeing. The .99 doesn't seem to be constrained by the actual known win rate of the top two BPP's whereas in the original discussion of the perfect handicapper they seemed to be restricted to the known win rate.

TD

[DeltaLover]

Hi Red Knave,
To be honest I didn't follow your post very well. My post was intended to point out that in order to have the number of winners in post # 45 (with the win rates shown in post # 35)
there would have needed to be almost 19000 races. Or, in the alternative, the win % would have needed to increase a good bit. Since the number of races shown was only 15000+ I was curious as to what we were seeing. The .99 doesn't seem to be constrained by the actual known win rate of the top two BPP's whereas in the original discussion of the perfect handicapper they seemed to be restricted to the known win rate.

TD

What we see is the following:
The number of SELECTED A type winners
The number of SELECTED B type winners
ALL OTHER WINNERS
By defualt we have a 58% correct rate A over B

[ReplayRandall]

What we see is the following:
The number of SELECTED A type winners
The number of SELECTED B type winners
ALL OTHER WINNERS
By defualt we have a 58% correct rate A over B


Delta, so bottom-line, we are trying to find a methodology that identifies the 42% correct rate B over A, while simultaneously not losing much of our 58% correct rate of A over B, thus getting as close to 100% proficiency as possible by choosing just one horse of top two BPP's......

[Light]

Delta, so bottom-line, we are trying to find a methodology that identifies the 42% correct rate B over A, while simultaneously not losing much of our 58% correct rate of A over B, thus getting as close to 100% proficiency as possible by choosing just one horse of top two BPP's......

134 posts just to get back to the original question (assuming it is the right question because I too am lost as to the question). No progress from point A regardless of what the question is. (LOL).

[DeltaLover]

Delta, so bottom-line, we are trying to find a methodology that identifies the 42% correct rate B over A, while simultaneously not losing much of our 58% correct rate of A over B, thus getting as close to 100% proficiency as possible by choosing just one horse of top two BPP's......

I think we reached this conclusion several pages ago

[ReplayRandall]

I think we reached this conclusion several pages ago


You got my point! We haven't progressed one iota for the last 2-3 pages.........That Deadline idea I had should have been implemented, but what do I know.......

[DeltaLover]

You got my point! We haven't progressed one iota for the last 2-3 pages.........That Deadline idea I had should have been implemented, but what do I know.......

I need some ideas... Do you have any? Are there any other things we should invastigate about this A X B concept? Any suggestions? ( by the way I hate deadlines and agile project management :) )

[Red Knave]

My post was intended to point out that in order to have the number of winners in post # 45 (with the win rates shown in post # 35)As I mentioned there is a transposition error in the 58% figures. The PP2 wins should read 1794 not 1974.

The .99 doesn't seem to be constrained by the actual known win rate of the top two BPP's whereas in the original discussion of the perfect handicapper they seemed to be restricted to the known win rate.I can't speculate on how DL got his numbers. I can see that some of them are suspect but the important ones, the 31% strike rate and break-even ROI are close enough to satisfy me.
With a 52% win rate the average mutuel, in the 100% example, would have been about $6.58 if the ROI was 1.72. If everything else except the average mutuel happened at 58% then the win rate should be 30.3% and the ROI would be about .99.
Basically if you could decide which one of PP1 or PP2 would prevail over the other and got it right about 60% of the time, you would be a winner.

[DeltaLover]

As I mentioned there is a transposition error in the 58% figures. The PP2 wins should read 1794 not 1974.

I can't speculate on how DL got his numbers. I can see that some of them are suspect but the important ones, the 31% strike rate and break-even ROI are close enough to satisfy me.
With a 52% win rate the average mutuel, in the 100% example, would have been about $6.58 if the ROI was 1.72. If everything else except the average mutuel happened at 58% then the win rate should be 30.3% and the ROI would be about .99.
Basically if you could decide which one of PP1 or PP2 would prevail over the other and got it right about 60% of the time, you would be a winner.

The decision (A or B) was done based on a 58% prefabricated level. By this I mean that the simulator is guaranteed to have this edge albeit with no regards to the proper rate of A over B

[TexasDolly]

As I mentioned there is a transposition error in the 58% figures. The PP2 wins should read 1794 not 1974.

I can't speculate on how DL got his numbers. I can see that some of them are suspect but the important ones, the 31% strike rate and break-even ROI are close enough to satisfy me.
With a 52% win rate the average mutuel, in the 100% example, would have been about $6.58 if the ROI was 1.72. If everything else except the average mutuel happened at 58% then the win rate should be 30.3% and the ROI would be about .99.
Basically if you could decide which one of PP1 or PP2 would prevail over the other and got it right about 60% of the time, you would be a winner.

Hi Red Knave,
It is more complicated than that since BPP #1 already wins at 60 % rate of those races that either BPP1 or BPP2 win and in that case would not have a breakeven roi. There has to be
consideration of the roi as well. The BPP #1 wins 30% and BPP #2 wins 20 % of the overall total according to the stats posted by DL.

TD

[Red Knave]

It is more complicated than that since BPP #1 already wins at 60 % rate of those races that either BPP1 or BPP2 win and in that case would not have a breakeven roi.
I just don't understand what you are saying here. I know that PP1 wins 50% more often than PP2 but I don't see how it affects the ROI other than what I have already calculated above.
The average mutuel is for all the winners, whether PP1 or PP2. It won't change that dramatically regardless of whether we get it right 100% or 58% of the time. What changes is the ROI.

The decision (A or B) was done based on a 58% prefabricated level. By this I mean that the simulator is guaranteed to have this edge albeit with no regards to the proper rate of A over BI actually guessed that you used a (semi)random number generator function and took a winner if the generated number was <= .58 and the winner was (PP1 or PP2). The number of winners you posted in the 58% PP2 column (1974 in post 45) is still too high, though, even for that. Test it for yourself. The ratio should not change that much. It should still remain 1.5 or so.

[TexasDolly]

I just don't understand what you are saying here. I know that PP1 wins 50% more often than PP2 but I don't see how it affects the ROI other than what I have already calculated above.
The average mutuel is for all the winners, whether PP1 or PP2. It won't change that dramatically regardless of whether we get it right 100% or 58% of the time. What changes is the ROI.

I actually guessed that you used a (semi)random number generator function and took a winner if the generated number was <= .58 and the winner was (PP1 or PP2). The number of winners you posted in the 58% PP2 column (1974 in post 45) is still too high, though, even for that. Test it for yourself. The ratio should not change that much. It should still remain 1.5 or so.

Hi Red Knave,
I wasn't saying much really except that if you just wanted to
win 60 % of the races you only need to pick the BPP #1 every race since it wins 60 % of the races decided by either BPP #! or BPP #2. Unfortunately, it doesn't produce a + ROI.
TD

[DeltaLover]

I actually guessed that you used a (semi)random number generator function and took a winner if the generated number was <= .58 and the winner was (PP1 or PP2). The number of winners you posted in the 58% PP2 column (1974 in post 45) is still too high, though, even for that. Test it for yourself. The ratio should not change that much. It should still remain 1.5 or so.

The logic goes like this:



if (winner is one of the top two):
if (random < 0.58)
if winner is A
increase A winners
else
increase A winners


Note that the analogy A:B ~ 1.5 is not assured... This is one of the tricky parts of this approach and one of the main associated challenges, where we can focus our attention.. All other topics we discussed so far (like GT, use the odds in our selection etc) are not directly related to the problem..

[Light]

Don't you think the answer to post #1 is doing a study of PP1 and PP2 to understand which one to choose based on form?

For example, take a PP1 who has been off over 100 days, dropping in class and has 1 workout 3 weeks ago. Obviously the horse still has problems. Out. In the same race you have a PP2 who just missed vs same 2 weeks ago. In. Its a no brainer which of these top 2 PP's will beat the other here. Of course many times it isn't so obvious. But isn't form factor going to have the most weight as to which PP will beat the other. Seems obvious to me.

[TexasDolly]

The logic goes like this:



if (winner is one of the top two):
if (random < 0.58)
if winner is A
increase A winners
else
increase A winners


Note that the analogy A:B ~ 1.5 is not assured... This is one of the tricky parts of this approach and one of the main associated challenges, where we can focus our attention.. All other topics we discussed so far (like GT, use the odds in our selection etc) are not directly related to the problem..

Hi DL,
The clause "if the winner is one of the top two " will have the BPP1/BPP2 ratio = 1.5 since that is the ratio they exist
in the data(from your stat post # 35). Red KNave wondered if the 1974 had been transposed from 1794 and it seems if that
might be the case. The ratio would then become 2711/1794 ~ 1.51. I would think any large number of observations of top two winners would be in that ratio regardless of the % of times you handicap them properly.
TD

[DeltaLover]

Hi DL,
The clause "if the winner is one of the top two " will have the BPP1/BPP2 ratio = 1.5 since that is the ratio they exist
in the data(from your stat post # 35). Red KNave wondered if the 1974 had been transposed from 1794 and it seems if that
might be the case. The ratio would then become 2711/1794 ~ 1.51. I would think any large number of observations of top two winners would be in that ratio regardless of the % of times you handicap them properly.
TD

At this moment I do not have access the code and db that created the data. I will post an update on this tomorrow.

[Red Knave]

Don't you think the answer to post #1 is doing a study of PP1 and PP2 to understand which one to choose based on form?
The PP number already includes form.
I think DL is looking for something he can apply without being subjective.

[ReplayRandall]

Funny thought, if you want 100% accuracy from top 2 BPP's per post #1, then those two horses would have to be a coupled entry......:D

[TrifectaMike]

This might help or not!

Let's look at a finite sequence of length, k. For demonstration purposes, let k = 20.

One way to proceed. First a histogram table would be generated showing the probability of each binary sequence occuring in the sequence.

Histogram table (0 -> pp2 wins, 1 -> pp1 wins; wins is definied as a heads-up match between pp1 and pp2))

Sequence Probability

0 .............. P(pp2 wins) --- root state or model 1
1 .............. P(pp1 wins)
00 ............ P(pp2 wins)*P(pp2 wins)
01 .............P(pp2 wins)*P(pp1 wins)
11 .............P(pp1 wins)*P(pp1 wins)
000 ...........P(pp2 wins)*P(pp2 wins)*P(pp2 wins)
001 ...........P(pp2 wins)*P(pp2 wins)*P(pp1 wins)

and so on.

These can be viewed as states or models.

Precious Mike

[Wizard of Odds]

Perhaps, more promising is to try to separate favorite from second favorite.

The first two favorites have a higher win rate than top two BRIS or top two ML.

In order to incorporate odds, I suggest maximizing actual minus expected wins (incorporating takeout). Of course, one can only expect to achieve a profit if the discrimination criteria between first and second favorite is so good that it overcomes the takeout. This is much easier said than done!

Yes, I know that it is sometimes difficult to ascertain the first two favorites before betting closes.

Just my two cents.


Wizard

[lansdale]

This might help or not!

Let's look at a finite sequence of length, k. For demonstration purposes, let k = 20.

One way to proceed. First a histogram table would be generated showing the probability of each binary sequence occuring in the sequence.

Histogram table (0 -> pp2 wins, 1 -> pp1 wins; wins is definied as a heads-up match between pp1 and pp2))

Sequence Probability

0 .............. P(pp2 wins) --- root state or model 1
1 .............. P(pp1 wins)
00 ............ P(pp2 wins)*P(pp2 wins)
01 .............P(pp2 wins)*P(pp1 wins)
11 .............P(pp1 wins)*P(pp1 wins)
000 ...........P(pp2 wins)*P(pp2 wins)*P(pp2 wins)
001 ...........P(pp2 wins)*P(pp2 wins)*P(pp1 wins)

and so on.

These can be viewed as states or models.

Precious Mike

Hi TM,

Keeping in mind my math-challenged nature, I have to say this looks suspiciously like the Bayesian version of the sequence test of the pattern-seeking Shannon-machine example i was requesting from DL, while imagining a frequentist model. True?

One possible negative, which I mentioned - my examples, Shannon, poker, involved only individual opponents. Although one opponent might display such predictably patterned behavior, would the pattern disappear as crowd size increases, and the 'wisdom of crowds' e.g. the betting public, kicks in? Even given this limitation, it seems possible that smaller tracks, with fewer bettors, might still display a distinctive profile of non-randomized sequence betting.

Cheers,

lansdale

[lansdale]

Poki-bot? :cool:

Since it sounds like you've visited their site, you know that they've developed a number of bots (Poki was one) at least one of which was able to compete against high-level pros. As far as the poker 'long run' research I mentioned, I'm not sure whether this was simmed or not. But I think any gambler can take a clue from the magnitude of the variance.

[JJMartin]

Perhaps, more promising is to try to separate favorite from second favorite.

The first two favorites have a higher win rate than top two BRIS or top two ML.

In order to incorporate odds, I suggest maximizing actual minus expected wins (incorporating takeout). Of course, one can only expect to achieve a profit if the discrimination criteria between first and second favorite is so good that it overcomes the takeout. This is much easier said than done!

Yes, I know that it is sometimes difficult to ascertain the first two favorites before betting closes.

Just my two cents.


Wizard

Although true, there would not be enough time to process a race at post time not to mention that invariably the #1 fav cannot be determined with accuracy until after the race is run.

[Magister Ludi]

PP Ranking Occurrences Winners Act % Win Expected % Win
1////////////14890////////4695////31.53//////38.84
2////////////14890////////3076////20.66//////24.24
3////////////14890////////2113////14.19//////16.91
4////////////14864////////1606////10.80//////12.08
5////////////14661////////1210/////8.25////////8.41
6////////////13821/////////856//////6.19////////5.57
7////////////11876/////////530//////4.46////////3.40
8/////////////9148//////////345//////3.77////////1.95
9/////////////6322//////////218//////3.45/////////.99
10////////////4106//////////136//////3.31/////////.46
11////////////1967///////////51//////2.59//////////.18
12/////////////991///////////38///////3.83//////////.06
13///////////////84///////////0////////0.00//////////.01
14///////////////40///////////0////////0.00//////////.00

[ReplayRandall]

PP Ranking Occurrences Winners Act % Win Expected % Win
1////////////14890////////4695////31.53//////38.84
2////////////14890////////3076////20.66//////24.24
3////////////14890////////2113////14.19//////16.91
4////////////14864////////1606////10.80//////12.08
5////////////14661////////1210/////8.25////////8.41
6////////////13821/////////856//////6.19////////5.57
7////////////11876/////////530//////4.46////////3.40
8/////////////9148//////////345//////3.77////////1.95
9/////////////6322//////////218//////3.45/////////.99
10////////////4106//////////136//////3.31/////////.46
11////////////1967///////////51//////2.59//////////.18
12/////////////991///////////38///////3.83//////////.06
13///////////////84///////////0////////0.00//////////.01
14///////////////40///////////0////////0.00//////////.00

ML, the win rate table you posted is not reliable because it is not specific enough for EXACT field sizes.

[Magister Ludi]

ML, the win rate table you posted is not reliable because it is not specific enough for EXACT field sizes.

I have calculated for the field sizes. E.g., there are (40) 14-horse races. There are (44) 13-horse races: (84) total 13th-ranked horses minus the (40) 13th-ranked horses in the 14-horse races.

[ReplayRandall]

I have calculated for the field sizes. E.g., there are (40) 14-horse races. There are (44) 13-horse races: (84) total 13th-ranked horses minus the (40) 13th-ranked horses in the 14-horse races.
This was my post(#69) to Delta on Nov.27th:
The chart is not a true representation of what each Prime power rank win rate is, against a specific number of horses in a field. For instance, we are handicapping a race of 12 starters, so by looking at the chart, we take out the 40 starts of 14 horse fields and 84 starts of 13 horse fields. Now we take 14,890-(40+84)=14,766, and that is now the current sample for all races of at least 12 horse fields. However, we do not know the number of wins PP1 had when we threw out those 120 races, thus skewing the rest of the table. The win rate is not reliable because it is not specific enough for EXACT field sizes. Other posters have mentioned this dilemma and I am convinced they are correct...........more filtering is needed.
Delta understood and replied with the following:

http://themindofagambler.com/top-bpp.jpeg

My questions to you: Do you believe this is a worthwhile exercise even it is impossible to satisfy Delta's Post #1 question? Are you in agreement with Sapio's post? What ideas need to be implemented now, in order to boost this thread to the next level?

[Wizard of Odds]

You really only need to know top 2 Odds regardless of order. You can also run scenarios in ambiguous cases which may all result in elimination of same horse if, for example, you can only really tell 2 of 3 horses will be first and second in betting with 1 minute to post.

Also, if your algorithm is automated, it won't matter that much that you enter data with only a minute or two minutes to post.

Best,
Wizard

[Magister Ludi]

This was my post(#69) to Delta on Nov.27th:

Delta understood and replied with the following:

http://themindofagambler.com/top-bpp.jpeg

My questions to you: Do you believe this is a worthwhile exercise even it is impossible to satisfy Delta's Post #1 question? Are you in agreement with Sapio's post? What ideas need to be implemented now, in order to boost this thread to the next level?

I haven't been following this thread. I just saw the easy money in Mr. Dl's original chart and it caught my attention. Mr. Dl's chart referenced in your link is for only the top Bris PP pick. I haven't read Mr. Sapio's post(s).

Just do a simple visual inspection of the winners % column. Doesn't it strike you as very "bottom-heavy"? My odds line applied to Mr. Dl's original chart shows that Bris seems to be expertly picking mid to long shots.

[ReplayRandall]

I haven't been following this thread. I just saw the easy money in Mr. Dl's original chart and it caught my attention. Mr. Dl's chart referenced in your link is for only the top Bris PP pick. I haven't read Mr. Sapio's post(s).

Just do a simple visual inspection of the winners % column. Doesn't it strike you as very "bottom-heavy"? My odds line applied to Mr. Dl's original chart shows that Bris seems to be expertly picking mid to long shots.

Yes, I see the same anomalies. If true, it looks like profit could be extracted with minimal effort, but small sample sizes may have skewed the win percentages. ML, any further expansion of thought, by you, on this thread would be appreciated......

[TrifectaMike]

Hi TM,

Keeping in mind my math-challenged nature, I have to say this looks suspiciously like the Bayesian version of the sequence test of the pattern-seeking Shannon-machine example i was requesting from DL, while imagining a frequentist model. True?

One possible negative, which I mentioned - my examples, Shannon, poker, involved only individual opponents. Although one opponent might display such predictably patterned behavior, would the pattern disappear as crowd size increases, and the 'wisdom of crowds' e.g. the betting public, kicks in? Even given this limitation, it seems possible that smaller tracks, with fewer bettors, might still display a distinctive profile of non-randomized sequence betting.

Cheers,

lansdale

People are not very good at choosing a purely random sequence.

"Shannon-machine example" essentially exploits the fact that people do not behave as "randomly" as they think they do.

I believe what DL is attempting to do is binary prediction (as in Gambler's fallacy, hot hand, etc). What is the next 'value" in a sequence. So, he can match a sequence "better" than random chance.

How well it can be done is questionable. In my post I suggested a starting point. Basically a Markov Process Model (prior states and sampling from an urn with replacement) or a Bayesian approach (Models).

Precious Mike

[DeltaLover]

At this moment I do not have access the code and db that created the data. I will post an update on this tomorrow.

For documentation purposes the code that generated the results:

import card
import random
total_winnings, total_races, total_wins = 0.0, 0, [0, 0]

for race in card.race_stream(max_size=20000, verbose=True, load_results_from_db=True):
if race.winner:
total_races += 1
if race.winner.prime_power_ranking in (1, 2):
if random.random() <= 0.58:
total_winnings += race.winner.win_payoff
total_wins[race.winner.prime_power_ranking - 1] += 1

total_bet = total_races * 2.
print 'Total Amount Bet: {} Total Winnings: {} ROI: {:2.2f} '.format(total_bet, total_winnings, total_winnings / (total_races * 2.))
print 'total races: ', total_races
print 'A: {} B: {} C: {}'.format(total_wins[0], total_wins[1], total_races - sum(total_wins))
print 'A / B = {:2.2f}'.format(total_wins[0] * 1.0 / total_wins[1])


Since a randomizer is used, I cannot verify the data as appear in the picture. Running the code again gives the following:

Total Amount Bet: 29748.0 Total Winnings: 30137.6 ROI: 1.01
total races: 14874
A: 2748 B: 1800 C: 10326
A / B = 1.53

Same program for p <= 0.5

Total Amount Bet: 29748.0 Total Winnings: 25337.3 ROI: 0.85
total races: 14874
A: 2313 B: 1539 C: 11022
A / B = 1.50

[TexasDolly]

For documentation purposes the code that generated the results:

import card
import random
total_winnings, total_races, total_wins = 0.0, 0, [0, 0]

for race in card.race_stream(max_size=20000, verbose=True, load_results_from_db=True):
if race.winner:
total_races += 1
if race.winner.prime_power_ranking in (1, 2):
if random.random() <= 0.58:
total_winnings += race.winner.win_payoff
total_wins[race.winner.prime_power_ranking - 1] += 1

total_bet = total_races * 2.
print 'Total Amount Bet: {} Total Winnings: {} ROI: {:2.2f} '.format(total_bet, total_winnings, total_winnings / (total_races * 2.))
print 'total races: ', total_races
print 'A: {} B: {} C: {}'.format(total_wins[0], total_wins[1], total_races - sum(total_wins))
print 'A / B = {:2.2f}'.format(total_wins[0] * 1.0 / total_wins[1])


Since a randomizer is used, I cannot verify the data as appear in the picture. Running the code again gives the following:

Total Amount Bet: 29748.0 Total Winnings: 30137.6 ROI: 1.01
total races: 14874
A: 2748 B: 1800 C: 10326
A / B = 1.53

Same program for p <= 0.5

Total Amount Bet: 29748.0 Total Winnings: 25337.3 ROI: 0.85
total races: 14874
A: 2313 B: 1539 C: 11022
A / B = 1.50

Hi Dl,
Thanks for verifying my point that the a/b ratio should stay around 1.5.
TD

[DeltaLover]

Hi Dl,
Thanks for verifying my point that the a/b ratio should stay around 1.5.
TD

Yes, you made a good point!

Still, the rate is not constant. It depends on how many times we are going to be correct.. For example if we assume that we will only be correct by 20% then the rate can go higher:

Total Amount Bet: 29748.0 Total Winnings: 10370.0 ROI: 0.35
total races: 14874
A: 956 B: 610 C: 13308
A / B = 1.57

[TexasDolly]

Hi Dl,
A variation yes, but it's not big really. We are now faced with
finding another condition which might help lead us to the 58 % breakeven point. Unfortunately, my thinking always heads toward parameters which undoubtedly, have been considered over and over in the original BPP calculation.

TD

[Light]

The PP number already includes form.


What I meant by "form" was the form of the horse coming into today's race since his last race.There is no numerical rating for that period of time and it is up to the handicapper to determine whether that form will affect his performance today in a positive,negative or neutral way.

I think DL is looking for something he can apply without being subjective.

I think you're right.

[DeltaLover]

So far, we have only consider limiting our selections between the top and second prime power.

Is there anything magic about these two? What is going to be if instead of the 1 - 2 we decide to use some other pair?

It can be 1 - 5, 2 - 3 or whatever else. For example if we use 4 - 5, we will see that the breakpoint is going to be 0.66:


Total Amount Bet: 29748.0 Total Winnings: 30173.0 ROI: 1.01
total races: 14874
A: 1058 B: 825 C: 12991
A / B = 1.28

Is it valid at this point, to assume that for each pair (1-2 or 4-5 or 2 - 9 or whatever) there is a breakpoint threshold?

If this is correct, can we assume that we can concentrate in any race in finding a specific pair of horses where we have a specific rating of success that we try to exceed?

[raybo]

I'm sorry I'm late to this thread, but after reading the first page of posts, I got tired of reading them.

Your original hypothesis was between the 1st and 2nd ranked BPP horses, in each race, at all tracks. You want to know (I think anyway, very confusing the way it is stated, to me) how many times you must hit the winner to break even, in ROI, if you have a 100% effective way of selecting the winner from those 2 horses. Whew!

Then you posted a table showing the long term (I assume) hit percentages of the BPP rankings. The 1st BPP ranked horse, in all those races, has a hit percentage of 31.53%. The 2nd ranked BPP horse has a hit percentage of 20.66%. So, the absolute maximum percentage of winners using those 2 horses is 52.19%. That is the very best you can do, if you choose the correct horse, of those 2, 100% of the time.

So, your break even point, regarding ROI hinges on the average payout you receive over the whole sample of races in which one of those 2 BPP horses actually wins. So, if betting a $2 flat wager each time, in 14,890 races you would bet $29,780. You will hit 52.19% of all races which would be 7771.091 winners. You will automatically receive your bet back for all of those winners which would be $15,542.182. Subtract that from the total amount bet and you have $14237.818 that you need to get from those horses' odds.

Am I missing something else? Or, has the original answer been provided already? I'm not going to go back and read all the other posts in this thread right now, so excuse me if this has been provided already.

[classhandicapper]

I think this problem would be a lot easier to solve if you had the average payoff of the top ranked horse and average payoff for the second ranked horse. Then you could calculate how much you needed to increase the win percentage of the 1st ranked horse to put it in the black by moving some of the 2nd ranked horses that won to the top (assuming the average payoff remained steady when you shifted the rank). Presumably you would be able to do that because you are a little smarter at ranking the top 2 horses than BRIS.

If you are talking about continuing to bet the top 2 ranked horses, then you would need more payoff information about the 3rd or even 4th ranked horses because again you are theoretically smart enough to put them into the top 2 and improve the overall win% of the top 2.

[Red Knave]

What about just randomly betting PP1 60% of the time and PP2 40%?

[thaskalos]

So far, we have only consider limiting our selections between the top and second prime power.

Is there anything magic about these two? What is going to be if instead of the 1 - 2 we decide to use some other pair?

It can be 1 - 5, 2 - 3 or whatever else. For example if we use 4 - 5, we will see that the breakpoint is going to be 0.66:


Total Amount Bet: 29748.0 Total Winnings: 30173.0 ROI: 1.01
total races: 14874
A: 1058 B: 825 C: 12991
A / B = 1.28

Is it valid at this point, to assume that for each pair (1-2 or 4-5 or 2 - 9 or whatever) there is a breakpoint threshold?

If this is correct, can we assume that we can concentrate in any race in finding a specific pair of horses where we have a specific rating of success that we try to exceed?

DL, forgive me for taking this conversation a little outside the parameters that you've set in this thread...but I've had this question ever since I first saw this thread of yours:

It's obvious that you've done some research on these Prime Power numbers...and it's also obvious that these numbers have shown an admirable predictive ability. It's a noteworthy achievement when a methodology's top-rated choice wins more races than its second choice, who in turn wins more races than the third choice...and so on, all the way down the line. Might there be a "weakness" in these numbers that we might be able to identify...so we could lend a helping hand and perhaps boost them into the profitability level?

I have been able to do something similar with the Beyer speed figures...but I knew in advance what the methodology behind these figures was, so I knew where the improvement was needed. Is it possible to do the same with the Prime Power Figures?

[raybo]

What about just randomly betting PP1 60% of the time and PP2 40%?

Just my admittedly lacking knowledge of stats, etc., but wouldn't your results also be random? You still need some way of "deciding" when to bet the 1st and 2nd BPP horses, in order to achieve the 60/40 ratio. And, what seems obvious, you would many times be betting the 60% top BPP horses on many of the wrong races (randomly wrong), and vice versa, the 40%.

[TrifectaMike]

Raybo, read post 91.

Precious Mike

[raybo]

Raybo, read post 91.

Precious Mike

Ok, so we have already tossed out the hypothetical 100% effectiveness of the selection system between BPP#1 and BPP#2, as being impossible (obviously impossible).

So, now we need a selection system between BPP#1 and BPP#2 that produces somewhere between 1.0 ROI and 1.70ROI.

0.0 ROI would mean 0% efficiency in the "improved" selection system (0 hit rate), and 1.70 ROI would be 100% efficiency in the "improved" selection system (100% hit rate)? Half of that is 0.95 ROI, so our mimicking system needs to be slightly better than 50% efficient in order to break even, which is almost a random selection system, right? If so, then our improved selection system doesn't need to be too good in order to break even (0.051% improvement needed)? Or, am I missing something again?

[TexasDolly]

Ok, so we have already tossed out the hypothetical 100% effectiveness of the selection system between BPP#1 and BPP#2, as being impossible (obviously impossible).

So, now we need a selection system between BPP#1 and BPP#2 that produces somewhere between 1.0 ROI and 1.70ROI.

0.0 ROI would mean 0% efficiency in the "improved" selection system (0 hit rate), and 1.70 ROI would be 100% efficiency in the "improved" selection system (100% hit rate)? Half of that is 0.95 ROI, so our mimicking system needs to be slightly better than 50% efficient in order to break even, which is almost a random selection system, right? If so, then our improved selection system doesn't need to be too good in order to break even (0.051% improvement needed)? Or, am I missing something again?

Hi Raybo,there are a number of examples DL has run. It takes a correct selection rate of 58% to breakeven using BPP1 and BPP2, 66% for bpp4 and BPP5 as I recall. He has posted them as recently as today.
TD

[raybo]

Hi Raybo,there are a number of examples DL has run. It takes a correct selection rate of 58% to breakeven using BPP1 and BPP2, 66% for bpp4 and BPP5 as I recall. He has posted them as recently as today.
TD

So, I'm seeing 55.1% needed and he's showing 58% needed, what am I doing wrong?

[raybo]

Maybe I'm just dense, but if you have 2 horses to choose from, and if you always choose correctly then you will produce a 1.7 ROI. On the flip side, if you never hit a winner after "improving" the BPP, then you will produce a 0.0 ROI. if you flip a coin for BPP#1 and BPP#2 you will be right 50% of the time, which according to his initial post of 1.7 ROI with a 100% rate of choosing correctly, should produce a 0.95 ROI. You only need 5.000.....1 ROI points better than a dice throw to reach 1.000......1 ROI, right?

If so, then you only need to find out what factors are not included in the BPP and use those factors to produce a .051 ROI above what the flat BPP produces.(?)

[TrifectaMike]

Maybe I'm just dense, but if you have 2 horses to choose from, and if you always choose correctly then you will produce a 1.7 ROI. On the flip side, if you never hit a winner after "improving" the BPP, then you will produce a 0.0 ROI. if you flip a coin for BPP#1 and BPP#2 you will be right 50% of the time, which according to his initial post of 1.7 ROI with a 100% rate of choosing correctly, should produce a 0.95 ROI. You only need 5.000.....1 ROI points better than a dice throw to reach 1.000......1 ROI, right?

If so, then you only need to find out what factors are not included in the BPP and use those factors to produce a .051 ROI above what the flat BPP produces.(?)

Raybo, read post #55

Precious Mike

[raybo]

Yeah, I gotta be missing something, because it sure looks like it wouldn't be that hard to improve a dice throw by 5+%. And, if it isn't then why hasn't someone done it already? Or have they?

[DeltaLover]

Maybe I'm just dense, but if you have 2 horses to choose from, and if you always choose correctly then you will produce a 1.7 ROI. On the flip side, if you never hit a winner after "improving" the BPP, then you will produce a 0.0 ROI. if you flip a coin for BPP#1 and BPP#2 you will be right 50% of the time, which according to his initial post of 1.7 ROI with a 100% rate of choosing correctly, should produce a 0.95 ROI. You only need 5.000.....1 ROI points better than a dice throw to reach 1.000......1 ROI, right?

If so, then you only need to find out what factors are not included in the BPP and use those factors to produce a .051 ROI above what the flat BPP produces.(?)

You are a little confused...

As I have said befor, if you are correct 50% the results look like this:

Same program for p <= 0.5

Total Amount Bet: 29748.0 Total Winnings: 25337.3 ROI: 0.85
total races: 14874
A: 2313 B: 1539 C: 11022
A / B = 1.50

ROI is similar to betting always the top of the second horse, something that it is exactly we should expect.. It turns out that only a 6% increase in your correct picks, turns the 0.85 to a fair bet with zero EV...

[raybo]

Raybo, read post #55

Precious Mike

Ok, Sapio says that BPP#1 should win 64% of the time, and DL says 1.5, or 60% of the time. Which is correct? The rest is above my capacity.

[raybo]

You are a little confused...

As I have said befor, if you are correct 50% the results look like this:

Same program for p <= 0.5

Total Amount Bet: 29748.0 Total Winnings: 25337.3 ROI: 0.85
total races: 14874
A: 2313 B: 1539 C: 11022
A / B = 1.50

ROI is similar to betting always the top of the second horse, something that it is exactly we should expect.. It turns out that only a 6% increase in your correct picks, turns the 0.85 to a fair bet with zero EV...

Your 6%, versus my 5.1% not much difference IMO. Even though I didn't have to write a program to find that out. The question remains, if only a 5-6% increase over a coin toss will put you in the black, why hasn't it been done already, if it hasn't.

And, I am a bit more than "a little confused". This techno stuff just does not compute to me.

[DeltaLover]

Your 6%, versus my 5.1% not much difference IMO. Even though I didn't have to write a program to find that out. The question remains, if only a 5-6% increase over a coin toss will put you in the black, why hasn't it been done already, if it hasn't.

And, I am a bit more than "a little confused". This techno stuff just does not compute to me.

I do not really think there is a lot of techno stuff in this thread.. Please let me know if you need some clarifications, as a concept, what we try to do here is very simple. If you are confused, this it means that I have not done a good job describing my thoughts, which again are very simple...

Note that I do not pretend that what I am saying here, is any kind of a silverbullet "system" that can make you any kind of money.. As I have repeated many times, I do not believe that something like this exist.. I am merely trying to analyze a beting approach to a components that might reveal some helpfull concepts that can be uses as accros the board generic guidelines...

[DeltaLover]

Ok, Sapio says that BPP#1 should win 64% of the time, and DL says 1.5, or 60% of the time. Which is correct? The rest is above my capacity.

I do not know where you found the 60% of the time.. I think we said that

The cumulative winning frequencies for BPP are as follows:

http://www.themindofagambler.com/bpp-winning-freq.jpeg


While the winning frequency of the top PP by field size can be seen here:

http://themindofagambler.com/top-bpp.jpeg

[ReplayRandall]

Are there any other things we should investigate about this A X B concept? Any suggestions?

Since we're using BRIS Prime Power, how about using other BRIS categories to filter BPP1/BPP2 such as: Average Class last 3, Current Class, Back Speed, etc. We may find inversions or anomalies that tilt the %'s more toward the objective.....

[DeltaLover]

Since we're using BRIS Prime Power, how about using other BRIS categories to filter BPP1/BPP2 such as: Average Class last 3, Current Class, Back Speed, etc. We may find inversions or anomalies that tilt the %'s more toward the objective.....

Absolutely. Once we have a clear methodology in hand, it will be trivial to extend it to any kind of a ranking.

[raybo]

I do not know where you found the 60% of the time.. I think we said that

The cumulative winning frequencies for BPP are as follows:

http://www.themindofagambler.com/bpp-winning-freq.jpeg


While the winning frequency of the top PP by field size can be seen here:

http://themindofagambler.com/top-bpp.jpeg


Someone in one of the threads mentioned a 60/40 ratio. That's where the 60% came from.

What is confucing me is that, if the ROI is evenly distributed, then you stated that you would have 1.7ROI at 100% efficiency, and I stated that then you would have 0.0ROI at 0% efficiency, and then, exactly in the middle lies 50% efficiency, and 0.95 ROI. Obviously, because neither you not Sapio agree then the distribution is not an even one. Otherwise, a coin toss between BPP#1 and BPP#2 would give you a 50% efficiency of 0.95ROI, and you would only need to improve that by 5+ points to be profitable.

[ReplayRandall]

Absolutely. Once we have a clear methodology in hand, it will be trivial to extend it to any kind of a ranking.

Ok...Let's start by isolating just those races where BPP2 finished in front of BPP1. Start the filtering process of other BRIS categories and see what singular or group of categories signify the strongest signal(most common factors) in %'s of why BPP2 topped BPP1. Now reverse process to just those races BPP1 tops BPP2, and filter again to find strongest signal. Finally, input selections, per TM's post #91 string of 1,2,2,1,1,2,1 etc. of 14,890 races and record results.

[Wizard of Odds]

I guess I'm not following...

Does any of the work in this thread lead to doing better than a .85 ROI (i.e., losing 15%)?

If one wanted a ranking than seems to perform similarly to BRIS rating, then Morning Line ranking seems to perform similarly (at decent sized tracks). However, while the win rate is progressively higher with progressively lower morning lines, the ROI will still virtually always be .90 or lower. Unless you are able to find some subset of overlays (very difficult), BRIS and Morning Lines don't lead to profitability by themselves.

Am I missing something?

[DeltaLover]

I guess I'm not following...

Does any of the work in this thread lead to doing better than a .85 ROI (i.e., losing 15%)?

If one wanted a ranking than seems to perform similarly to BRIS rating, then Morning Line ranking seems to perform similarly (at decent sized tracks). However, while the win rate is progressively higher with progressively lower morning lines, the ROI will still virtually always be .90 or lower. Unless you are able to find some subset of overlays (very difficult), BRIS and Morning Lines don't lead to profitability by themselves.

Am I missing something?

I agree that discovering the subsets you are talking about, seems like the obvious way to improve this (or any other) model... Again the purpose of this thread is not necessary to find a profitable approach as it can reveal a component that can be used as a signal to a more advanced solution..

[ReplayRandall]

Ok...Let's start by isolating just those races where BPP2 finished in front of BPP1. Start the filtering process of other BRIS categories and see what singular or group of categories signify the strongest signal(most common factors) in %'s of why BPP2 topped BPP1. Now reverse process to just those races BPP1 tops BPP2, and filter again to find strongest signal. Finally, input selections, per TM's post #91 string of 1,2,2,1,1,2,1 etc. of 14,890 races and record results.


Quick clarification; Finally, input selections, per TM's post #91 string of 1,2,2,1,1,2,1 etc. of 14,890 different/new set races and record results.

[raybo]

I guess I'm not following...

Does any of the work in this thread lead to doing better than a .85 ROI (i.e., losing 15%)?

If one wanted a ranking than seems to perform similarly to BRIS rating, then Morning Line ranking seems to perform similarly (at decent sized tracks). However, while the win rate is progressively higher with progressively lower morning lines, the ROI will still virtually always be .90 or lower. Unless you are able to find some subset of overlays (very difficult), BRIS and Morning Lines don't lead to profitability by themselves.

Am I missing something?

BPP is highly correlated to the final odds, so if you accept that you will not know what the final odds are, then BPP would probably be the best you could do, as far as predicting the final odds. That is about the only thing that I can see that would be beneficial from this experiment. It still comes down to hit rate versus average price, when you're talking about being profitable. Yes, we all attempt to improve what we are currently doing, or have done in the past, but regardless of how you approach it, it all comes down to those 2 things. You must improve one without sacrificing the other, or vice versa. And that, is no small thing.

[Wizard of Odds]

It still comes down to hit rate versus average price, when you're talking about being profitable. Yes, we all attempt to improve what we are currently doing, or have done in the past, but regardless of how you approach it, it all comes down to those 2 things. You must improve one without sacrificing the other, or vice versa. And that, is no small thing.

You have said it all!

The only tiny modification I would make (and which I know you meant anyway) is that you can sacrifice some win rate if you gain profitability (e.g., long term 20% win rate with 1.2 ROI would be a huge success and very difficult to attain, whereas long term 34% win rate with .8 ROI is failure for us and easy to attain).

[pandy]

Does anyone know what the ROI is for, say, the first 5 Bris Prime Power ranked horses? Is the ROI higher for the first two, or is it higher for horses ranked third, or fourth?

In my experience, the best way to produce a positive ROI betting to win on thoroughbreds is to have a hit rate of anywhere between 15 and 25%. A hit rate of 33% or so is very difficult to profit from because that would mean that you are playing too many favorites.

[Wizard of Odds]

BPP is highly correlated to the final odds, so if you accept that you will not know what the final odds are, then BPP would probably be the best you could do, as far as predicting the final odds. .

Perhaps, Morning Line beats BPP for win rate at big handle tracks?

For NY and Cal circuit tracks, I think it's a horse race :D between the two in terms of win rate. (Could use program selections to break ties in Morning Line). Of course, used alone and betting every race, both BPP and Morning Line will be unprofitable.

[DeltaLover]

Does anyone know what the ROI is for, say, the first 5 Bris Prime Power ranked horses? Is the ROI higher for the first two, or is it higher for horses ranked third, or fourth?


It is already posted several times in this thread:


http://www.themindofagambler.com/bpp-winning-freq.jpeg

[pandy]

These are the win percentages, not the ROI, right?

[DeltaLover]

These are the win percentages, not the ROI, right?

Here you can see the ROI as well:


http://themindofagambler.com/bris-pp-roi.jpeg

[classhandicapper]

I still think you guys may be missing an important part of this (unless I missed some things earlier in the thread).

Assume you have found a flaw in the Bris ratings. That flaw has allowed you to improve your selection ability between the #1 ranked horse and #2 ranked horse by x%, but the average payoff of the extra winners is $4.00. x% will have to be fairy large for you to turn a profit.

Now assume the same things, except the extra winners are paying an average of $10. x% can be a lot smaller.

Without knowing what the average payoff of the extra winners is going to be, you can't know how much you have to increase you win percentage to get to profit.

Within a ROI of .86 (or whatever) for the 2nd ranked horses is a mix of a lot of different priced winners. Moving a few of them up to the 1st rank won't be random. You are WAY more likely to find flaws within any automated rating among the lower priced horses in the group. The same would be true if you were trying to move 3rd and 4th ranked horses into the top 2 so you could bet both.

[raybo]

Perhaps, Morning Line beats BPP for win rate at big handle tracks?

For NY and Cal circuit tracks, I think it's a horse race :D between the two in terms of win rate. (Could use program selections to break ties in Morning Line). Of course, used alone and betting every race, both BPP and Morning Line will be unprofitable.

The M/L is also reflected in the final odds, as many people use those M/L odds as one man's determination of the worth of the horses (that of course depends on whether the odds maker is making probability assessments or attempting to predict the way the final odds will end up). The final odds contain all assessment methods of all bettors in the win pool, including those who use the M/L odds in their assessment.

[raybo]

I still think you guys may be missing an important part of this (unless I missed some things earlier in the thread).

Assume you have found a flaw in the Bris ratings. That flaw has allowed you to improve your selection ability between the #1 ranked horse and #2 ranked horse by x%, but the average payoff of the extra winners is $4.00. x% will have to be fairy large for you to turn a profit.

Now assume the same things, except the extra winners are paying an average of $10. x% can be a lot smaller.

Without knowing what the average payoff of the extra winners is going to be, you can't know how much you have to increase you win percentage to get to profit.

Within a ROI of .86 (or whatever) for the 2nd ranked horses is a mix of a lot of different priced winners. Moving a few of them up to the 1st rank won't be random. You are WAY more likely to find flaws within any automated rating among the lower priced horses in the group. The same would be true if you were trying to move 3rd and 4th ranked horses into the top 2 so you could bet both.

Of course that is true. But, this exercise is supposed to be about improving the discrimination between the top ranked and 2nd ranked BPP numbers. It has already been stated that if you could correctly discriminate between those 2 horses (100% of the time) your ROI would be 1.7. The question is how do you go about more closely approaching that 100% rate, or at least improving it enough to reach profitability.

[Wizard of Odds]

The M/L is also reflected in the final odds, as many people use those M/L odds as one man's determination of the worth of the horses (that of course depends on whether the odds maker is making probability assessments or attempting to predict the way the final odds will end up). The final odds contain all assessment methods of all bettors in the win pool, including those who use the M/L odds in their assessment.

Agreed!

I was just trying to point out that while Delta has shown about 52% win rate with top two BPP, the top two ML (at NY and CA) is going to be near that and might even slightly beat it?

Of course, top two ML will also have ROI's about .86 ish

[raybo]

Agreed!

I was just trying to point out that while Delta has shown about 52% win rate with top two BPP, the top two ML (at NY and CA) is going to be near that and might even slightly beat it?

Of course, top two ML will also have ROI's about .86 ish

Yes, individual tracks will perform differently using any rating or method you can imagine. That is the nature of the game, tracks don't all play the same. That's why I personally believe that any method you use should be tailored to, at least, track.

[DeltaLover]

Agreed!

I was just trying to point out that while Delta has shown about 52% win rate with top two BPP, the top two ML (at NY and CA) is going to be near that and might even slightly beat it?

Of course, top two ML will also have ROI's about .86 ish


Here you can see how morning line is behaving from a win % and ROI perspective:

Ranked based on MORNING LINE


http://themindofagambler.com/morning-line-rankings.jpeg

[Wizard of Odds]

Here you can see how morning line is behaving from a win % and ROI perspective:

Ranked based on MORNING LINE


http://themindofagambler.com/morning-line-rankings.jpeg

Thanks ....So win rate goes up from 52% to 54% and ROI goes down from .86 to .83

[classhandicapper]

Of course that is true. But, this exercise is supposed to be about improving the discrimination between the top ranked and 2nd ranked BPP numbers. It has already been stated that if you could correctly discriminate between those 2 horses (100% of the time) your ROI would be 1.7. The question is how do you go about more closely approaching that 100% rate, or at least improving it enough to reach profitability.

If that's the question (which it didn't seem to be earlier in the thread) then you would want to know how the figures are generated and look for flaws in the process/methodology for creating them or the way the horses are ranked.

That's one way I search for value. I've tried to understand the methodology of all the leading figure makers the best I can and then try to poke holes in what they are doing. If I can find just a handful of situations where the major figure makers are faulty, I should be able to outperform those figures by a small amount.

[DeltaLover]

Here you can see the same metrics for FINAL ODDS


http://themindofagambler.com/final-odds-rankings.jpeg

[raybo]

If that's the question (which it didn't seem to be earlier in the thread) then you would want to know how the figures are generated and look for flaws in the process/methodology for creating them or the way the horses are ranked.

That's one way I search for value. I've tried to understand the methodology of all the leading figure makers the best I can and then try to poke holes in what they are doing. If I can find just a handful of situations where the major figure makers are faulty, I should be able to outperform those figures by a small amount.

Trying to find flaws in the BPP is going to be very, very tough, because it includes so many weighted factors, and Brisnet is not going to reveal all the factors and none of their weightings. If you knew what factors were not included in the BPP, then maybe you could use those to improve the discriminations between 1st and 2nd ranked BPP horses.

[Wizard of Odds]

Here you can see the same metrics for FINAL ODDS


http://themindofagambler.com/final-odds-rankings.jpeg

Kinda blows your mind that almost 3/4 winners are in lowest 3 odds. Partially as a result of field size reduction I know..., BUT STILL!

[DeltaLover]

Kinda blows your mind that almost 3/4 winners are in lowest 3 odds. Partially as a result of field size reduction I know..., BUT STILL!

Still, besides the high strike rating, you can never show a profit by following the crowd. This is why we should not try to pick winners but large(r) prices.

[thaskalos]

Kinda blows your mind that almost 3/4 winners are in lowest 3 odds. Partially as a result of field size reduction I know..., BUT STILL!

This might be the time to implement a strategy similar to the one that William L. Scott wrote about in his book "Investing at the Racetrack"...where betting was confined to the three shortest odds in the race.

[raybo]

This might be the time to implement a strategy similar to the one that William L. Scott wrote about in his book "Investing at the Racetrack"...where betting was confined to the three shortest odds in the race.

I believe this is what this whole thread is about. How to improve your ability to distinguish between the top 2 or 3 rankings, regardless of the rankings method used.

[thaskalos]

I believe this is what this whole thread is about. How to improve your ability to distinguish between the top 2 or 3 rankings, regardless of the rankings method used.

This is what I've been thinking about too lately. If you can't beat them...join them.

[raybo]

This is what I've been thinking about too lately. If you can't beat them...join them.

Well, it still boils down to hit rate and average price, so you have to make a decision on which is more important, and at what mix ratio do you combine them.

[DeltaLover]

I believe this is what this whole thread is about. How to improve your ability to distinguish between the top 2 or 3 rankings, regardless of the rankings method used.

The specific case we are dealing with here, is just representative of a more generic problem. Whenever we are representing the relative strength of the horses in a race with some kind of a ranking (regardless of whether we are using BPP, M/L or anything else), the handicapping problem will boil down to some kind of a selection process resembling the one that is presented in this thread.

[thaskalos]

The specific case we are dealing with here, is just representative of a more generic problem. Whenever we are representing the relative strength of the horses in a race with some kind of a ranking (regardless of whether we are using BPP, M/L or anything else), the handicapping problem will boil down to some kind of a selection process resembling the one that is presented in this thread.

I'm sorry DL...I am trying to get over a bad cold these last few days, and my "focus" isn't as sharp as it would otherwise be. What "selection process" has been presented in this thread?

[davew]

I'm sorry DL...I am trying to get over a bad cold these last few days, and my "focus" isn't as sharp as it would otherwise be. What "selection process" has been presented in this thread?

the selection process where you can take the top 2 BPP, and pick the winner between them 100% of the time to get 1.7 ROI and 60% overall wins

maybe I am missing the part about skipping 80%+ races.... and then getting it

[PICSIX]

I apologize if this is slightly off topic....I believe one's time is better spent developing a power ranking of their own versus some sort of improvement to BPP. The problem with BPP is EVERYONE has it!

***Warning*** The following is from memory (I will search for the exact info after I post this):

Jeff Platt has/had a rating he created in J-Capper that had a positive ROI flat betting the top ranked horse until he released it to a few hundred customers. After that, the positive ROI was gone. When too many have the same info it becomes less useful.

[Wizard of Odds]

This might be the time to implement a strategy similar to the one that William L. Scott wrote about in his book "Investing at the Racetrack"...where betting was confined to the three shortest odds in the race.


Just decided to, quite mindlessly, look at winners of first five races at Turf Paradise today....

All 5 from top 2 final odds....

[PICSIX]

I apologize if this is slightly off topic....I believe one's time is better spent developing a power ranking of their own versus some sort of improvement to BPP. The problem with BPP is EVERYONE has it!

***Warning*** The following is from memory (I will search for the exact info after I post this):

Jeff Platt has/had a rating he created in J-Capper that had a positive ROI flat betting the top ranked horse until he released it to a few hundred customers. After that, the positive ROI was gone. When too many have the same info it becomes less useful.

My memory was slightly off. Scroll down to the question, "Why did you take the original JRating enabled program off the market?"

http://www.jcapper.com/faqs.asp

[thaskalos]

the selection process where you can take the top 2 BPP, and pick the winner between them 100% of the time to get 1.7 ROI and 60% overall wins

maybe I am missing the part about skipping 80%+ races.... and then getting it

If you can pick the winner of the top two picks of ANY "respectable" methodology, 100% of the time...aren't you odds-on to show a profit in the end? Couldn't the same be done with the tote board top two choices?

[Sapio]

Hi all,

222 posts and no forward movement.

Any ideas?

Thomas Sapio

[Wizard of Odds]

Hi all,

222 posts and no forward movement.

Any ideas?

Thomas Sapio


Hilarious!

Find a discrimination function that separates first and second favorite (incorporating odds) by using another 'reliable' index besides odds (e.g., BPP index)

Not sure if this qualifies as a new idea:bang:

[MJC922]

Considering only the top two of a comprehensive ratings method that's optimized for profit (or at least to minimize loss) is not a bad way to go about things IMO as anything below that rank ought to be losing substantial money overall. Merely being ranked below two others that have more 'profitable' things going for them quickly becomes more of a negative than a positive so third rank down is typically a waste.

To choose one or the other then probably ought to come down to looking at what doesn't go into the ratings themselves. That could be anything which prime power doesn't bring into their model. I don't know what BRIS is using so it's difficult to say, but certainly anything like race day tote action or body language is probably going to be value-added info if someone is skilled in knowing what to look for.

[raybo]

I'm sorry DL...I am trying to get over a bad cold these last few days, and my "focus" isn't as sharp as it would otherwise be. What "selection process" has been presented in this thread?

LOL. I believe the thread has not gotten that far yet. But, thus far, it's been about "process" rather than individual selection criteria.

[raybo]

the selection process where you can take the top 2 BPP, and pick the winner between them 100% of the time to get 1.7 ROI and 60% overall wins

maybe I am missing the part about skipping 80%+ races.... and then getting it

That certainly could be part of the process, but then you would probably destroy the 60% hit rate, wouldn't you.

[pandy]

Still, besides the high strike rating, you can never show a profit by following the crowd. This is why we should not try to pick winners but large(r) prices.


I agree. The fastest way to a profit using the BPP would be to figure out how to get the win percentage around 20% while only betting longshots.

[raybo]

I apologize if this is slightly off topic....I believe one's time is better spent developing a power ranking of their own versus some sort of improvement to BPP. The problem with BPP is EVERYONE has it!

***Warning*** The following is from memory (I will search for the exact info after I post this):

Jeff Platt has/had a rating he created in J-Capper that had a positive ROI flat betting the top ranked horse until he released it to a few hundred customers. After that, the positive ROI was gone. When too many have the same info it becomes less useful.

I believe that he had a power rating that had a slightly lower hit rate than the BPP, but produced a "slightly" better ROI, meaning it was still a negative ROI, not a flat bet profitable one.

[Light]

Hi all,

222 posts and no forward movement.

Any ideas?

Thomas Sapio

There is no answer here because this is just a case of a carrot dangling in front of the handicapper. If there was an answer it would have been discovered years before this thread. I've been down this road before in my 30 years as a player and went over this question with my database and found no single factor or multiple factors as to why PP1 beats PP2 or vice versa. Same for why first or second choice beats the other. Only handicapping sheds light on this and then its no longer a system because the factors are too large to create a "system". On top of that, anyone reading this who knows the answer will not post it. But I doubt anyone knows because my database shows each of the top 2 Bpp contradicting themselves as to what factors win.

[hardboot]

We can easily filter these reports by any additional conditions (like classification, distance or surface).. Still, this approach belongs to lower levels of handicapping that are not part of the macro-handicapping I am performing here. At this point, I am trying to keep things as simple as possible, in other words: KISS

just an off the wall thought from the cheap seats
and
not to come off topic too far ....

I suppose a single parameter (such as distance) would provide a different PP1 vs PP2 hit rate from your posted all races chart, would it not.

I've no idea how BRIS arrives at the PP rating but I feel there should be vastly differing weighted options for arrival of a PP1 and a PP2 concerning a sprint vs a route

[donmat]

Whatever kind of power rating that's used lately I'm finding that post time odds is the most important factor for ROI. If I can get even money or better than I show a profit, anything below that the profit goes away. This could be a problem especially at small tracks when the horse is even money going in to the gate, but when the race goes off the horse is 3-5

[DeltaLover]

just an off the wall thought from the cheap seats
and
not to come off topic too far ....

I suppose a single parameter (such as distance) would provide a different PP1 vs PP2 hit rate from your posted all races chart, would it not.

I've no idea how BRIS arrives at the PP rating but I feel there should be vastly differing weighted options for arrival of a PP1 and a PP2 concerning a sprint vs a route

Unfortunately it is not that simple.. Here you can see how the top 5 rankings are behaving based on distance:


http://themindofagambler.com/roi-by-distance.jpeg

[Wizard of Odds]

Unfortunately it is not that simple.. Here you can see how the top 5 rankings are behaving based on distance:


http://themindofagambler.com/roi-by-distance.jpeg



Nice Chart!

How about a matrix of BPP Rank (column), M/L Rank row ... displaying win % (ROI).

Thanks

[DeltaLover]

Nice Chart!

How about a matrix of BPP Rank (column), M/L Rank row ... displaying win % (ROI).

Thanks

Here you go:

http://themindofagambler.com/bpp-ml.jpeg


each pair contains win% - roi for the corresponding BPP - ML pair..

[classhandicapper]

Trying to find flaws in the BPP is going to be very, very tough, because it includes so many weighted factors, and Brisnet is not going to reveal all the factors and none of their weightings. If you knew what factors were not included in the BPP, then maybe you could use those to improve the discriminations between 1st and 2nd ranked BPP horses.

Then IMO this thread is a waste of time.

I could easily make a list of specific circumstances where there are almost always huge discrepancies between the major figure makers. Someone is clearly getting it wrong and impacting the board in an exploitable way. They have their top 2 misranked in those situations. That's what you are looking for.

[Wizard of Odds]

Here you go:

http://themindofagambler.com/bpp-ml.jpeg


each pair contains win% - roi for the corresponding BPP - ML pair..

Thanks,

Well done!!! Very Interesting Indeed

How do you break ties in rankings to classify tied horses slotted into say M/L rank 2 or 3 when both have same M/L?

[DeltaLover]

Thanks,

Well done!!! Very Interesting Indeed

How do you break ties in rankings to classify tied horses slotted into say M/L rank 2 or 3 when both have same M/L?

Did not pay special attention to this case. When something like this happens, I re-sort by BPP

[Wizard of Odds]

Did not pay special attention to this case. When something like this happens, I re-sort by BPP

Highest BPP is the tiebreaker?

[DeltaLover]

Highest BPP is the tiebreaker?
yes

[Wizard of Odds]

yes

Are BPP's ever tied?

[DeltaLover]

Are BPP's ever tied?

Νο, BPP are never tied

[Johnny V]

Νο, BPP are never tied
There are ties in the BPP and I believe they do occur often enough. Just in todays races at LRL there are ties in the 1st, 6th and 9th races. In some other instances the difference between the PP number is so small that I wonder if there is any meaningful difference between them at that point.

[DeltaLover]

There are ties in the BPP and I believe they do occur often enough. Just in todays races at LRL there are ties in the 1st, 6th and 9th races. In some other instances the difference between the PP number is so small that I wonder if there is any meaningful difference between them at that point.

Yes, my statement is wrong. Indeed there are some (few) cases where two horses might have the same BPP... In a random sample of 1,000 races the ties are as follows:

total_races: 1000

Position : 2 Ties: 4
Position : 5 Ties: 3
Position : 4 Ties: 2
Position : 3 Ties: 1
Position : 7 Ties: 1

In the first spot ties are even more rare and when they occur, in most of the cases it is due to scratching the top horses...

Still, these discrepancies have not an impact to what we try to do here..

[raybo]

Yes, my statement is wrong. Indeed there are some (few) cases where two horses might have the same BPP... In a random sample of 1,000 races the ties are as follows:

total_races: 1000

Position : 2 Ties: 4
Position : 5 Ties: 3
Position : 4 Ties: 2
Position : 3 Ties: 1
Position : 7 Ties: 1

In the first spot ties are even more rare and when they occur, in most of the cases it is due to scratching the top horses...

Still, these discrepancies have not an impact to what we try to do here..

Well, it appears that there would be a degree of impact on the results, if the 2nd ranked horse is tied with the 3rd ranked BPP. Because, theoretically, the BPP#1 horse is in competition with 2 horses, the 2nd and 3rd ranked horses, and one of those 2 is not even being considered in the separation between BPP#1 and BPP#2.

IMO, all ties should be resolved, in some way, before attempting to decide which of a pair of rankings should be the final selection, even if it is only something as simple as post position.

[DeltaLover]

Well, it appears that there would be a degree of impact on the results, if the 2nd ranked horse is tied with the 3rd ranked BPP. Because, theoretically, the BPP#1 horse is in competition with 2 horses, the 2nd and 3rd ranked horses, and one of those 2 is not even being considered in the separation between BPP#1 and BPP#2.

IMO, all ties should be resolved, in some way, before attempting to decide which of a pair of rankings should be the final selection, even if it is only something as simple as post position.

The impact is minimal, and I do not want to spend time resolving them. Anyone with a different opinion should go ahead and resolve them and try to detect and report any discrepancies.

[Wizard of Odds]

Well, it appears that there would be a degree of impact on the results, if the 2nd ranked horse is tied with the 3rd ranked BPP. Because, theoretically, the BPP#1 horse is in competition with 2 horses, the 2nd and 3rd ranked horses, and one of those 2 is not even being considered in the separation between BPP#1 and BPP#2.

IMO, all ties should be resolved, in some way, before attempting to decide which of a pair of rankings should be the final selection, even if it is only something as simple as post position.

There are ties in odds ranking too and I'm not sure how DL handled them. While seemingly tied favorite is often flagged with an *, I've often seen other horses tied in odds on charts, e.g., two horses at odds of 4.20. Unless one had final win pool dollars bet, where a tie is much more unlikely, ties in odds are not uncommon at all.

[DeltaLover]

There are ties in odds ranking too and I'm not sure how DL handled them. While seemingly tied favorite is often flagged with an *, I've often seen other horses tied in odds on charts, e.g., two horses at odds of 4.20. Unless one had final win pool dollars bet, where a tie is much more unlikely, ties in odds are not uncommon at all.

These ties cannot change the large image, not at least to a significant manner.. Neither the ROIs nor the WIN % will change much any way of tie resolution you might use.

[raybo]

These ties cannot change the large image, not at least to a significant manner.. Neither the ROIs nor the WIN % will change much any way of tie resolution you might use.

I understand what you're saying, and generally that is true, unless one or more of those were huge prices, then they could have significant impact on the results. Yes, if that were the case you might end up deleting those races, as "outliers", from your database.

[lansdale]

People are not very good at choosing a purely random sequence.

"Shannon-machine example" essentially exploits the fact that people do not behave as "randomly" as they think they do.

I believe what DL is attempting to do is binary prediction (as in Gambler's fallacy, hot hand, etc). What is the next 'value" in a sequence. So, he can match a sequence "better" than random chance.

How well it can be done is questionable. In my post I suggested a starting point. Basically a Markov Process Model (prior states and sampling from an urn with replacement) or a Bayesian approach (Models).

Precious Mike

Thanks for clarifying this TM. It looks like DL has not taken you up on this, and that he and everyone else in the thread are looking for a solution based on fundamentals. Given the 15 years or so exposure of this figure, I can't see how it can happen. However, you seem doubtful that the type of pattern-mimicry DL is seeking can be achieved by any means.

Cheers,

lansdale