Playing Around With Prime Power Numbers

[reckless]

Robert, you're in my prayers and all of us here at PA are with you totally.

[sligg]

Originally Posted by Robert Goren

This is what I am playing with now. I working to determine how often the highest rated prime power horse beats a lower rated one by the difference in rating assuming at one of them won the race. I am not interest how they do if one runs 7th and the other ninth. I am only inter
ested if one of the two won. I now create a % for how often a horse beats every other horse in the race. Sort like Chess ratings. I do this for every horse in the race. Then using Bayes(thank you Trifecta Mike) I have a % win chance for each starter. I then use square root of Kelly to determine the size of the bet. I plug 75% of my % win chance for the estimate chance of winning in the Kelly formula. I am being very conservative because it so early in the development.
This is still in the very early stage of development and I am pretty sure the numbers I use for determining how much the difference wins will change some as I had more races to my sample. Right I am looking at straight predictor, but it maybe a log. Who Knows what I will eventual end up with. Right now I have dumped all dirt races between 5 1/2 f to a mile & 1/16 together assuming that BRIS has incorporated distance into their ratings. At some point I will separate routes from sprints if this show enough promise. I have compile all the data and enter into excel by hand so things move rather slowly. Throw in that I am bad health and it is a crawl. I did test run yesterday with a $50 bankroll at the AQU. Ended up $22 ahead.
I also plan to look at things that improve the prime powers like speed figures from a different source. But that right now is in the distant future. I will also look at the effects being overlayed or underlayed.
If my health gets to the point where I can no longer bet, I will share my numbers with anyone here who is interested.

Are you working with one track, Aqueduct?

Brisnet has a program Power Search Reports, which costs $6 per day that lists all the top PP horses with a 6 point minimum differential. Unfortunately at $6 per day (about $150 per month) is too expensive for me so I only use and bet Aqueduct using PP. But this gives very few plays.

If I accept the 6 point differential, then what you have is odds on favorites that win at a very high % with a very small ROI.

sligg@hotmail.com

[Robert Goren]

Originally Posted by sligg

Are you working with one track, Aqueduct?

Brisnet has a program Power Search Reports, which costs $6 per day that lists all the top PP horses with a 6 point minimum differential. Unfortunately at $6 per day (about $150 per month) is too expensive for me so I only use and bet Aqueduct using PP. But this gives very few plays.

If I accept the 6 point differential, then what you have is odds on favorites that win at a very high % with a very small ROI.

sligg@hotmail.com

I am current working with one track, Aqueduct because that what I have the most PPs for. I am not so much interested in the gap between the top horse and the second horse as I am about the gap between all the horses and how it predicts the chance of winning. The money may be in betting lower rated horses, but at this point I am not sure. I am of the belief that results at AQU will translate to other major track although there might have to be some minor adjustments for each track. At least that is my hope. As I have said before. I do not expect this to make money. I hope to get it close though.

[mikesal57]

Robert..

Hope his this helps you out...

This is on info going back to 7/2012
gap values - All dirt ( main and inner) ..and all dist for AQU


Mike

[mikesal57]

FYI..

heres the "ALL" button

[eqitec]

I've created the graph below in an attempt to simplify visually the relationships between horses' PPRs and the field average PPRs. I wanted a quick and dirty tool for use in horizontals when I've handicapped one or two races in the horizontal comprehensively but ran out of time for the others,

From experimentation thus far, value bets can be found from horses which tend towards positive variances without necessarily being the top one or two.
The graph below is for tomorrow's 5th at TAM.

[Robert Goren]

Thanks, mike for your numbers, they will give me something to check against. Equitec Your graph sort of shows what I am attempting to put in to numbers. I am trying to come up with a win % for each horse in the race based, at least at first, on their power numbers. It my hope that win % when combined with the odds will produce a fairly accurate assessment of each horse's chances.

[Robert Goren]

A bit of an update. My surgery was postponed because I came down with a case of bronchitis. It has not been rescheduled yet.

On the Prime Power numbers front. Sapio forwarded me some work done Trifecta Mike which include a formula for determining how often the top rated horse wins in a one on one match up. After checking it out in some samples I found TM's formula to very accurate. I will use it. By multiplying all the one on one match ups I come up with a % chance of a horse going unbeaten in them. Using Bayes, I then have a % for each horse winning the race. These numbers too look to be pretty accurate. The catch is that lower odds horses at a certain %level win more than higher odds horses. I am now looking to see if there is a way to a niche at each % level that will produce something resembling a profit. A made up example might look like this. At The Prime Power % level of 50% winners. A horse going off at 1/2 wins at rate of 75% showing a profit of $0.25 per $2.00 bet. Much work is left to be done. But I think one thing is probably becoming clear. This method will be on a lot of chalk horses. I still do not expect to find anything profitable.
Thanks to Sapio for forwarding TM's work and if someone is still touch with TM, please thank him for me.

[Sapio]

Originally Posted by Robert Goren

A bit of an update. My surgery was postponed because I came down with a case of bronchitis. It has not been rescheduled yet.

On the Prime Power numbers front. Sapio forwarded me some work done Trifecta Mike which include a formula for determining how often the top rated horse wins in a one on one match up. After checking it out in some samples I found TM's formula to very accurate. I will use it. By multiplying all the one on one match ups I come up with a % chance of a horse going unbeaten in them. Using Bayes, I then have a % for each horse winning the race. These numbers too look to be pretty accurate. The catch is that lower odds horses at a certain %level win more than higher odds horses. I am now looking to see if there is a way to a niche at each % level that will produce something resembling a profit. A made up example might look like this. At The Prime Power % level of 50% winners. A horse going off at 1/2 wins at rate of 75% showing a profit of $0.25 per $2.00 bet. Much work is left to be done. But I think one thing is probably becoming clear. This method will be on a lot of chalk horses. I still do not expect to find anything profitable.
Thanks to Sapio for forwarding TM's work and if someone is still touch with TM, please thank him for me.

Hi Robert

Hope you feel better.

Yes, I agree TM's formula is very accurate (so far everything I've tested that was made available by TM has been extremely accurate).

I'm presently using TM's Probability Generating Function to determine what percentage of the odds is driven by speed ratings.

Thomas Sapio

[Robert Goren]

When using a logit multi regression with PP and the odds , PP continues to be a small negative factor. That is the lower the PP is the more likely its the winner. But is a very small factor. I am now adding Trainer % and Jockey % on the running style. I have not gotten far enough along to have feel if either is going to be a factor.

[Robert Goren]

Small update. I have decide my R skills needed to be upgraded. Everything is on hold while I try to do that while fighting through my dizziness.

[socialnetworker]

Originally Posted by Robert Goren

Small update. I have decide my R skills needed to be upgraded. Everything is on hold while I try to do that while fighting through my dizziness.

Hi Robert: Sorry to hear you are not feeling well.
I have similar intentions to yourself: I belief more predictive power can be squeezed out of the Bris Ratings...

and Multiple Regression is definitely a level I would like to get to.

I am handicapped by not having complete archives of past Brisnet PPs.

I am working on a way to scan the contents of the BRIS Summary Reports into a database...

If anybody knows anybody who would like to contribute (or sell) their archived BRIS PPs for any of the big tracks, please let me know.

[Robert Goren]

BRIS Prime Power certainly has some predictive power. Whether there is a way to turn them into a useful tool is an open question. Of one thing I am sure, BRIS looked at everything in their PPs to come up with the Prime Power numbers. If you going to multiple regressions with it as a variable, you will need to look at variables that are not part of BRIS's PPs. Maybe Beyer's or CJ's speed ratings. Maybe something from the Sheets. I am pretty sure Jockey or Trainer stats are not in them. But such things as a general claimed in its last race is probably in them. My next project is to look at how the tote board odds effect the chances a horse winning at different Prime Power margins. But I need to learn a bit more about how to use R first.

[Robert Goren]

Good hunting, Socialnetworker

[Sapio]

Originally Posted by Robert Goren

When using a logit multi regression with PP and the odds , PP continues to be a small negative factor. That is the lower the PP is the more likely its the winner. But is a very small factor. I am now adding Trainer % and Jockey % on the running style. I have not gotten far enough along to have feel if either is going to be a factor.

Hi Robert

From a good source, try PP and 1 - 1/odds as regressors as a starting point.

odds = final tote odds

Thomas Sapio