2015 Saratoga Daily Variants

[cj]

Originally Posted by Cratos

Thanks; from your response I see your projections as probability multivariate projections.

I would agree with that.

[highnote]

steveb --- if you're still listening ---

Inyour experience is it better to find the average beyer deviation for all the races on a card whether they are sprints or routes, or is it better to mix them all together in order to get a bigger sample size?

For example, there are 3 six furlong races and 3 ten furlong races on the day. would you find the combined average deviation for all of them or would you find the average of the six furlong races and then find the average of the ten furlong races?

[steveb]

Originally Posted by highnote

steveb --- if you're still listening ---

Inyour experience is it better to find the average beyer deviation for all the races on a card whether they are sprints or routes, or is it better to mix them all together in order to get a bigger sample size?

For example, there are 3 six furlong races and 3 ten furlong races on the day. would you find the combined average deviation for all of them or would you find the average of the six furlong races and then find the average of the ten furlong races?

i don't do beyers highnote, mine are strictly my own, even if they were the inspiration.

i don't have any personal experience with your racing, so am not qualified to comment, except that i always uses all distances for myself wherever i have done it.

but i would first compensate for the way the race was paced before i would worry about the variant.
(and before dark target says anything, this is AFTER all that DT, when we already know the times and classes and whatnot)

or i might weight the races as to how much i think they contribute to the variant.
a race paced much slower than i would expect would get less weighting.
the faster the pace the higher the weighting.

so i would do those sorts of things before i decided if i needed a split variant or not.

but if you have the right method i can't personally see any reason why you need to split it between sprints and routes.

and of course my way will not be like yours or others ways so it may not be the right answer for you.

[highnote]

Originally Posted by steveb

i don't have any personal experience with your racing, so am not qualified to comment, except that i always uses all distances for myself wherever i have done it.

I have seen people normalize all the times and distances to one mile and then make the variants. I'm not sure it makes sense to normalize, though, because it might add more noise?

but i would first compensate for the way the race was paced before i would worry about the variant.

or i might weight the races as to how much i think they contribute to the variant.
a race paced much slower than i would expect would get less weighting.
the faster the pace the higher the weighting.

I have noticed, especially on grass route races, that the pace can be extremely slow in one race on the card and then much faster in the next race. You will see the same thing on dirt, but to a lesser extent. I don't know how you could make an accurate assessment of the speed of the surface under those conditions, unless you project the variant. So one concern I have about projecting the variant is that you move away from time based speed ratings into a method that resembles handicap ratings. However, if the handicap ratings produce more accurate predictions then the concern is unwarranted.

and of course my way will not be like yours or others ways so it may not be the right answer for you.

Maybe not, but I like your idea of analyzing the pace.

James Quinn writes about an interesting pace method in his book "Figure Handicapping" where he uses utilizes a "pace range" to rate the time of the final fraction in grass races. If my memory is correct, he says that if the pace is slow run, you would expect the final fraction to be quick, but the final time will suffer because there is not enough distance remaining in the race after the pace call for the horses to run a fast final time.

On the other hand, if the pace is too fast then the final fraction will tend to be slower which will cause the final time to be slower.

So if the pace time is run in the "sweet spot" range then the final fraction and final time will be fast.

Now, if the pace is slow and the final time is fast or if the pace is fast and the final time is fast then you might have an extraordinary horse.

Quinn creates a set of "pace range" pars.

In Beyer's most recent book, he discusses Quinn's method and points out some of the flaws as well as the good points.

Also, Cary Fotias wrote in his book "Blinkers Off" about a figure handicapping pattern called "Turf Decline Line" (C) where over the past three races the pace figure declines, but the final figure is about the same. This shows that the horse's closing speed is increasing because the horse is getting fitter.

I noticed yesterday at Saratoga the winners of the turf races all had "Turf Decline Line" over their past two or three races even though their final speed figures were not as high as others. And a given horse that had the highest pace and speed figure did not win. Whereas, on dirt, the horses with highest pace and final figures ran well.

[Tom]

When I see pace variants for, say, sprints, of

F5
F9
F7
S6
F8
F5

I just ignore the S6 completely.
Obviously, the track did not cause the slow pace, so why bother figuring it in?
I make the track F7 for the day, and the S6 race is now S13.

[highnote]

Originally Posted by Tom

When I see pace variants for, say, sprints, of

F5
F9
F7
S6
F8
F5

I just ignore the S6 completely.
Obviously, the track did not cause the slow pace, so why bother figuring it in?
I make the track F7 for the day, and the S6 race is now S13.

That makes sense.

[highnote]

Steveb -- if you're listening... are you familiar with Steven Brecher's book "Beating the Races with a Computer"?

He has a method of making speed ratings that looks interesting.

[steveb]

Originally Posted by highnote

Steveb -- if you're listening... are you familiar with Steven Brecher's book "Beating the Races with a Computer"?

He has a method of making speed ratings that looks interesting.

you rang!

never heard of him highnote.
why don't you critique it and i might have a read depending on if you reckon it's worthwhile or not.

i just searched for that book and it's from 1980!!
might be a bit dated?

[highnote]

Originally Posted by steveb

you rang!

never heard of him highnote.
why don't you critique it and i might have a read depending on if you reckon it's worthwhile or not.

i just searched for that book and it's from 1980!!
might be a bit dated?

Woods would have been familiar with the book and it's speed ratings as it is referenced in his former partner's paper.

I've read the details he spells out for making figures, but his process has not yet become clear to me.

You're right. The book is from the 1980s. There are probably better methods available that are easier to follow.

I'll try to write up a summary about the process for you.

[steveb]

Originally Posted by highnote

Woods would have been familiar with the book and it's speed ratings as it is referenced in his former partner's paper.

I've read the details he spells out for making figures, but his process has not yet become clear to me.

You're right. The book is from the 1980s. There are probably better methods available that are easier to follow.

I'll try to write up a summary about the process for you.

thank you.
so you are saying benter references it?

the only copy i could find was on abe books.
where i guess if i was still caring about racing, i would willingly buy it, if i thought it would help me, but i am too old now.
my time is done.
the jigsaw puzzle that is racing interests me still, but the racing itself bores me.

and, if i was honest, the only books that i thought worthwile, were the early beyer books, which got me on my way a zillion years ago.
have never seen a good one since those days.
not that i have read too many on times.
don scott and rem plante being best of the others, but they reckoned (wrongly) that time was not much good.
but their other stuff was good for somebody learning their way, as i was in the long ago.

a quick google say some nice things about him(brecher), although apparently wong is better according to some(on PA too)
i have the wong book and his speed stuff is colour by numbers.
some of wong's other stuff is a good read, but the times stuff is nonsense.

give me some clues highnote as regards brecher, you obviously have the book and think it ok.
you could always scan the relevant pages and email them to me, and i could then offer you some opinion/advice.

[highnote]

Brecher's approach to making speed ratings is interesting, but I think there are better ways.

First, he finds the final time of the race.

Second, he uses a multiple regression equation to predict the time of the race.

Third, he finds the difference between the actual time and his predicted time. He calls this difference the track variant. He sums the variant for all the races and then divides by the number of races to find the daily variant. The daily variant will be added to all the races on the day to create a final time that takes into account the speed of the track.

Fourth, he makes a Base Time for the race by setting all the factors in his multiple regression equation to a "detent" level. For example, if purse value is one of the multiple regression factors and it varies from $10,000 to $100,000 depending on the quality of the race, he will use a detent purse level of 10,000.

Fifth, he creates a standardized time by dividing the final time of the race plus the daily variant and dividing it by the base time. Then subtracts 1. This standardized time can be positive or negative where negative times are faster.

It's an interesting and unique approach, but I'm not convinced it is any better than any of the typical approaches to making speed figures.

[steveb]

Originally Posted by highnote

Brecher's approach to making speed ratings is interesting, but I think there are better ways.

First, he finds the final time of the race.

Second, he uses a multiple regression equation to predict the time of the race.

Third, he finds the difference between the actual time and his predicted time. He calls this difference the track variant. He sums the variant for all the races and then divides by the number of races to find the daily variant. The daily variant will be added to all the races on the day to create a final time that takes into account the speed of the track.

Fourth, he makes a Base Time for the race by setting all the factors in his multiple regression equation to a "detent" level. For example, if purse value is one of the multiple regression factors and it varies from $10,000 to $100,000 depending on the quality of the race, he will use a detent purse level of 10,000.

Fifth, he creates a standardized time by dividing the final time of the race plus the daily variant and dividing it by the base time. Then subtracts 1. This standardized time can be positive or negative where negative times are faster.

It's an interesting and unique approach, but I'm not convinced it is any better than any of the typical approaches to making speed figures.

bugger you highnote!!!
it is interesting when you say it like that.
i might have to buy a copy.

nothing unique about using regression to figure the time stuff, as i do it all the time.

your third paragraph is the correct way to do it imo, just that it needs more steps, and checks and balances which he probably had anyway.

can you enlarge on the bold bit??

and finally i had to get my dictionery out
'detent' , never heard the word before!

[highnote]

Originally Posted by steveb

can you enlarge on the bold bit??

and finally i had to get my dictionery out
'detent' , never heard the word before!

He calls his speed ratings "Standardized Normalized Times" or "SNT".

The bold bit describes in words a simple formula he shows:

          
      (Final Time + Variant)
SNT = ---------------------- - 1
            (Base Time)

He wrote:

"The first term in the formula may be thought of as a multiple of the time a mature male Base Purse class of horse would take to run the race. SNTs can be positive or negative. Negative is faster and approximates the logarithm of the ratio of the horse's time to the base time."

I think his approach is novel, but so is James Quinn's approach outlined in his book "Figure Handicapping". Quinn describes the best system I've seen for making speed ratings for grass races where horses are jockeyed for position early and then sprint to the wire -- as opposed to dirt races where the horses run as fast as they can for as far as they can and are decelerating as they approach the wire. As a general rule, on dirt, the horse that expends its energy optimally over the course of the race is the most likely winner.

Brecher's approach could be improved upon. He created his system sometime around 1980. It was designed to be an automated system where the figures were created by an automated computerized process. Nowadays, there are better ways to do this. Jerry Brown of Thorograph, and CJ of TimeformUS use a method where many horses from the same racecard are used to make projections rather than just using the winners' times.

Brecher uses just the winner's times and averages the individual race variants from the par time. His way is too general and misses the subtleties of a race. Plus, he does not take into account the pace of the race, whereas, James Quinn's method on grass racing does account for pace.

Have a look at the terrific race from 1993 called the Grade 1 Turf Classic Invitational Stakes run at Belmont Park on Long Island, not far from New York City at the link below. (scroll ahead to 8:50 for the start of the race) The winning horse was trained by Andre Fabre. Its name was Apple Tree. You might be familiar with the trainer and the horse.

The pace was dawdling for a 12 furlong grass race and the final time was 2:28. (For comparison, Secretariat ran 12 furlongs on the dirt in 2:24. Secretariat also ran 12 furlongs on turf in 2:24 and 4/5.)

With such a slow pace and final time how do you make an accurate speed rating for a race like the Turf Invitational? The horse on the lead in this race should have been able to win after setting such a slow pace. (For example, Buck's Boy set a slow pace when he won at 12 furlongs on grass in the 1998 Breeders Cup Classic.)

YouTube

[lansdale]

Thought it might be worth re-posting this venerable Benter piece, since Steve B is in this thread, but also to emphasize first, that the models created by himself and Woods were much more complex than this discussion implies, that they never relied on single figures, and that final time was only one among a huge array of handicapping factors.

Although it's clear he took some ideas for factor selection and defnition from Brecher, there's no suggestion here or anywhere in his writings that he relied on him for any overall analysis, as he did Bolton and Chapman. In any comparison between the ideas/techniques of Benter/Woods and Brecher, whose work was entirely theoretical, I would tend to rely on the former pair.

Cheers,

lansdale



https://www.scribd.com/doc/166556276/Benter

[highnote]

Originally Posted by lansdale

Thought it might be worth re-posting this venerable Benter piece, since Steve B is in this thread, but also to emphasize first, that the models created by himself and Woods were much more complex than this discussion implies, that they never relied on single figures, and that final time was only one among a huge array of handicapping factors.

Although it's clear he took some ideas for factor selection and defnition from Brecher, there's no suggestion here or anywhere in his writings that he relied on him for any overall analysis, as he did Bolton and Chapman. In any comparison between the ideas/techniques of Benter/Woods and Brecher, whose work was entirely theoretical, I would tend to rely on the former pair.

Cheers,

lansdale



https://www.scribd.com/doc/166556276/Benter

Agreed.

Speed figures help to predict race outcomes, but there are so many inaccuracies about them that you still need a comprehensive approach to handicapping in order to win.

However, there may be some people who can win using speed figures only. My guess is that in order to win using speed figures only they would have to be very selective of the types of races they bet. And even then they would probably have to handicap a little. I doubt that there are any speed figures that could be used blindly that could win consistently and produce a profit.