On the DRF's Speed Rating-Track Variant

[citygoat]

good call just couldnt get to 7

[betovernetcapper]

Originally Posted by cj

I thought I'd read everything he ever wrote but I don't remember that. Do you remember where he wrote it?

He wrote it in a book where he described living in Australia for a year. He was making figures & doing OK but recognized the difference. He also made a reference to the importance of the final fraction in one of his last 2 or 3 Webinars.

[betovernetcapper]

Originally Posted by DeltaLover

Since you disagree, I would like to know your answer to the following questions in regards to Beyer Speed figure creation ( I can ask many more questions but for now let's start from these)..

(1) What are the parallel time curves that are used?

Obviously what he states in his books are not only completely out of date but also do not cover difficult cases. The examples he provides refer to very soft cases, like for example six to seven furlongs on dirt but there is not a word about how to treat real world cases where we have very few races for a specific surface / distance.

(2) How cross - track adjustments are estimated?

He writes on "Beyer On Speed":



Does this mean that all southern Cal tracks are averaging to the same figure?
Does the same apply for northern? How shippers from other tracks are to be considered? How you handle shippers from a track who has very few or no shippers at all?

(3) Although Bayer claims that "class" represents a fictitious quality, he still uses it as the starting point for his figure creation, taking a specific classification as the base for his scale (He assigns the arbitrary number of 80 to the average 10K claimer). Is not this a contradiction to the "speed" approach?

(4) The value of a length in the Beyer methodology reflects the percentage worth of a fifth (moved by two decimal places) in comparison to the base (parallel) time of the race in fifths (converted to fifths, or in other words the figure is given by the following algorithm:

value_of_point = 1000. / (base_time * 5)
time_delta_in_fifths = base_time - time
points_delta = time_delta_in_fifths * value_of_point
figure = int(round(80. + points_delta))

How does he arrive in this formula and more precisely why he needs to use the full time of the race instead of a fraction of it (for example the last half of the race or something similar)? What is the impact of this approach as the distance becomes larger?

Let me make it clear that at no point I have the intention of downgrade the greatness of Beyer's work whom I consider the most influential handicapper ever and the patriarch of the American handicappers. I am just tring to make the case that his figures (that probably worked fine before they became part of the DRF) are antiquated and there is a lot of room for improvement for those who have the required knowledge and time to work on them.

As far as using par charts, he moved on to projected time a long time ago. Tracks change & he no doubt has made adjustments as needed. He uses the final time, because that is what his procedure is about. Other people make figures that include weight, ground lost, early or late speed but he doesn't. He makes a figure to reflect the final time. As far as using a different measure for beaten lengths, it make perfect sense. If a horse is beaten a length at 5 furlongs it's more significant then getting beat a length at a mile & a half.
If you have additional questions, I'd advise you contact Beyer directly
http://www.drf.com/news/andrew-beyer

[steveb]

Originally Posted by thaskalos

If Beyer has said that...then I've missed it. I've seen him use the final fraction and the Beyer figure in combination...but I haven't seen him comment on the importance of the final fraction alone. In fact, in his latest book, he suggested that the last HALF-MILE might be a better alternative.

if he did say it he was wrong.
sections(fractions) very important, but they should be used in tandem, one is not much good without the other.
the time is the speed, the fraction is the how.

as for the value of one fraction over another, then your data should tell you that answer, for each track, surface, and distance

[steveb]

Originally Posted by VigorsTheGrey

When the Daily Racing Form shows the speed figure-track variant numbers, for examples 89-08, 93-05, 102-01...

I get that the left figure is the speed rating; and the right figure is the track variant.

Am I correct that the higher the track variant the SLOWER the racing oval was that race...?

My question is: Does it make any sense to subtract the track variant from the speed figure, yes, of course....But what does that tell us...?

Also, what exactly is a 00 track variant...? Is 00 a super-fast surface or just a normally fast surface...?

I see that there are no minus (-04, example) track variants used to indicate a faster than normal surface...

The track variant must be tied to an optimum moisture content value for a 00 rating and yet no figure is used when the dirt is dried back, beyond that optimum rating...and therefore, faster than normal...

no idea of how others would do it.
for me.....
positive numbers are slow
negative numbers fast
zero is neutral


and no it makes no sense to subtract(nor add) variant from speed......to me

[VigorsTheGrey]

Originally Posted by steveb

no idea of how others would do it.
for me.....
positive numbers are slow
negative numbers fast
zero is neutral


and no it makes no sense to subtract(nor add) variant from speed......to me

I still don't really understand it myself, adding or subtracting the variant with the speed number...right now I'm going with what Augenj said to ADD them together to get an adjusted number...

...but a number of posters have remarked the these figures are antiquated and/or deeply flawed values and should not be used...which make me question why the are taking up space in the DRF form line if that is the case...

...still I have friends that base the whole handicapping regime around these figures with the Beyer number as well and are satisfied with them...

...but I think the track variant is already "built in" resulting in a slower or faster run time (with increased moisture levels slowing down the runners naturally) so in a way I can see your point about neither adding NOR subtracting the track variant...

...I just want to be clear about what I am looking at when I see a figure like 100-01, or 105-02...How high does the speed rating go...? I seldom see ones higher than 103...

I do like to use these figures along with the Beyer number but I'm not so clear about what to do with them...

[steveb]

Originally Posted by DeltaLover

Since you disagree, I would like to know your answer to the following questions in regards to Beyer Speed figure creation ( I can ask many more questions but for now let's start from these)..

(1) What are the parallel time curves that are used?

Obviously what he states in his books are not only completely out of date but also do not cover difficult cases. The examples he provides refer to very soft cases, like for example six to seven furlongs on dirt but there is not a word about how to treat real world cases where we have very few races for a specific surface / distance.

(2) How cross - track adjustments are estimated?

He writes on "Beyer On Speed":



Does this mean that all southern Cal tracks are averaging to the same figure?
Does the same apply for northern? How shippers from other tracks are to be considered? How you handle shippers from a track who has very few or no shippers at all?

(3) Although Bayer claims that "class" represents a fictitious quality, he still uses it as the starting point for his figure creation, taking a specific classification as the base for his scale (He assigns the arbitrary number of 80 to the average 10K claimer). Is not this a contradiction to the "speed" approach?

(4) The value of a length in the Beyer methodology reflects the percentage worth of a fifth (moved by two decimal places) in comparison to the base (parallel) time of the race in fifths (converted to fifths, or in other words the figure is given by the following algorithm:

value_of_point = 1000. / (base_time * 5)
time_delta_in_fifths = base_time - time
points_delta = time_delta_in_fifths * value_of_point
figure = int(round(80. + points_delta))

How does he arrive in this formula and more precisely why he needs to use the full time of the race instead of a fraction of it (for example the last half of the race or something similar)? What is the impact of this approach as the distance becomes larger?

Let me make it clear that at no point I have the intention of downgrade the greatness of Beyer's work whom I consider the most influential handicapper ever and the patriarch of the American handicappers. I am just tring to make the case that his figures (that probably worked fine before they became part of the DRF) are antiquated and there is a lot of room for improvement for those who have the required knowledge and time to work on them.

1....parralel time curves?????
would you not just need data to figure each track and distance?

2.....
the way to figure different jurisdicational strangths, is to use what you people refer to as the variant.
you just need to stat with the same baseline figures for each place of interest. the difference will be the difference in the variant.
i know that unless you know what i am talking about that this may be hard to grasp.
nevertheless i have been doing it for 30 or more years and it works fine, not that it's that important to know the difference between melbourne australia and hong kong for example.
but it's important to know the difference between sydney and melbourne or melbourne and provincial areas in the same state, or the difference between different jurisdictions that have high cross over points

3....the way to figure class is to make them ALL equal and then iterate and adjust.
trial and error
but class is just the collective race speed for that type, whereas the race speed is just the individual

4....there should be NO constant value for a length, because beaten 1 second on a heavy track is not the same as being beaten 1 second on a fast one.
beaten 1 second on turf here is not the same as beaten 1 second on turf there.
ditto dirt.
beaten 1 second in a fast time race is not the same as beaten 1 second in a slow time race.
it all about relativities.


but beyer was my inspiration in the long ago.
but i was probably different to most people that read his books.
i was not interested in the actual numbers, i wanted to know HOW he arrived at them.
once one understands the how, it is easy to see what errors are in his books.
but pretty sure the guy has earned the respect most people have for him though.

[steveb]

Originally Posted by betovernetcapper

As far as using par charts, he moved on to projected time a long time ago. Tracks change & he no doubt has made adjustments as needed. He uses the final time, because that is what his procedure is about. Other people make figures that include weight, ground lost, early or late speed but he doesn't. He makes a figure to reflect the final time. As far as using a different measure for beaten lengths, it make perfect sense. If a horse is beaten a length at 5 furlongs it's more significant then getting beat a length at a mile & a half.
If you have additional questions, I'd advise you contact Beyer directly
http://www.drf.com/news/andrew-beyer

why does it make sense?

a horse beaten a length over 5 furlongs, could very well be a better effort than one beaten 1 length over 1 and half miles.
what if they dawdled over the longer trip and were busting guts over the 5f?

your thinking is the same as post people on this subject, but i am confident it's wrong to think that way.

races are for the most part, dependent on what the leaders do.
how races are run is far more important than the distance of the race in this respect.

races are very rarely run as a race for the entire distance.
your 1.5 mile races may not be a race until they turn for home?
if they are not working then horses are very energy efficient apparently.

[Augenj]

Originally Posted by VigorsTheGrey

I still don't really understand it myself, adding or subtracting the variant with the speed number...right now I'm going with what Augenj said to ADD them together to get an adjusted number...

Here's why I do that for the speed component of my ratings. It's just one of eight components.

Let's say a horse has 2 races, one of them on a very slow track and the other on a very fast track. Both races are equal in class, distance, and surface. He finishes 1st in both of them.

Slow track - 50 speed + 50 variant = 100 adjusted speed
Fast track - 95 speed + 5 variant = 100 adjusted speed

It's obvious that he performs equally well on both slow and fast tracks. This is a gross simplification but can be used to get a slightly better (2 percent) win per cent if you average his adjusted speed figures over just averaging the speed figures without the variants.

The 2 percent comes from thousands of races at eleven major tracks over one year. That 2 percent value will vary by track and it's been a long time since I visited the stats behind it since speed is only a small piece of the THA ratings.

I also adjust the variant by multiplying it by 0.75 before combining it with the speed figure. This is an optimized figure from thousands of race calculations and is obvious when seen at the peak of a Bell Curve.

[DeltaLover]

Originally Posted by steveb

1....parralel time curves?????
would you not just need data to figure each track and distance?

2.....
the way to figure different jurisdicational strangths, is to use what you people refer to as the variant.
you just need to stat with the same baseline figures for each place of interest. the difference will be the difference in the variant.
i know that unless you know what i am talking about that this may be hard to grasp.
nevertheless i have been doing it for 30 or more years and it works fine, not that it's that important to know the difference between melbourne australia and hong kong for example.
but it's important to know the difference between sydney and melbourne or melbourne and provincial areas in the same state, or the difference between different jurisdictions that have high cross over points

3....the way to figure class is to make them ALL equal and then iterate and adjust.
trial and error
but class is just the collective race speed for that type, whereas the race speed is just the individual

4....there should be NO constant value for a length, because beaten 1 second on a heavy track is not the same as being beaten 1 second on a fast one.
beaten 1 second on turf here is not the same as beaten 1 second on turf there.
ditto dirt.
beaten 1 second in a fast time race is not the same as beaten 1 second in a slow time race.
it all about relativities.


but beyer was my inspiration in the long ago.
but i was probably different to most people that read his books.
i was not interested in the actual numbers, i wanted to know HOW he arrived at them.
once one understands the how, it is easy to see what errors are in his books.
but pretty sure the guy has earned the respect most people have for him though.

A few points to clarify my original question:

1. The problem with the parallel time creation lies in the fact that there exist many distances with rather unbalanced composition in race classifications; if all distances were containing the same mix of the various classifications (MSW – MCL – CLM- ALW – STK) the problem whould not exist, in the real world though this is not the case and needs a good solution to assure the accurateness of the produced figures. I believe that this issue is causing to large extend the problem of Beyer speed figures when it comes to turf and longer distances.

Here you can see a notebook where I show this problem using Saratoga turf:

https://github.com/deltalover/hoplat...analysis.ipynb

Note how unbalanced the breakdown is as we move from distance to distance. The legent of each piechart prints the average (raw) Beyer style figures having 100 as the base (instead of 80). The parallel time curve is created using Least Squares of the claiming times.


2. The intra-track variant is by no means a simple process as you write here. In this picture you can see a group of five different tracks, where x1, x2, .. x5 represent the cross -track adjustments while the average differences from track to track are represented with the deltas. The objective is to discover the best vector of x that minimizes the delta when applied to them. Considering that the real graph consists of over hundred of tracks and surfaces you understand that this is at least a complicated problem to solve.

[cj]

Some good stuff in this thread, amazing it started in a thread about a rating that is seriously flawed.

[betovernetcapper]

As long as this thread is exploring a number of things, did Beyer initially do speed figures for grass? I've been looking for my copy of My $50,000 year at the Races and can't find it. Quirin used to sell par charts every year & I think they were only for dirt.

[DeltaLover]

Originally Posted by betovernetcapper

As long as this thread is exploring a number of things, did Beyer initially do speed figures for grass? I've been looking for my copy of My $50,000 year at the Races and can't find it. Quirin used to sell par charts every year & I think they were only for dirt.

You might be right. I have heard Bayer (more than once) saying that he does not have much confidence to his figures on the grass. The reality is that if you do not have accurate turf figures in today's game you are going to pass way too much races which I see as as significant disadvantage.

[cj]

Originally Posted by betovernetcapper

As long as this thread is exploring a number of things, did Beyer initially do speed figures for grass? I've been looking for my copy of My $50,000 year at the Races and can't find it. Quirin used to sell par charts every year & I think they were only for dirt.

I think grass ratings came along later on. He didn't do them initially. Back when that book was written few tracks ran more than one race a day on turf.

[steveb]

Originally Posted by DeltaLover

You might be right. I have heard Bayer (more than once) saying that he does not have much confidence to his figures on the grass. The reality is that if you do not have accurate turf figures in today's game you are going to pass way too much races which I see as as significant disadvantage.

is there a difference.
they are both figured the same for me.
the relativities are different for sure, and the way they tire may be different, but so too are the relationships between races on heavy turf and fast turf.

thanks too, for your previous reply, but it's obvious to me we approach things in different ways.
who is right?
maybe both of us?

to me everything i need is figured by iteration and trial and error.
so that once the error can't be reduced, then it must be as accurate as it can be?