Speed Figures Revisited

[Aerocraft67]

TFUS for me. I don't have much experience comparing speed figs or doing my own. I suspect TFUS figs are the best because of the stature they've gained here and at DRF, and it seems like the most modern approach. That they translate to overseas races is a nice bonus.

Consider it a qualitative assessment as a data shopper rather than a scientific evaluation of their competitive accuracy. I think the DRF marketing of "exclusive" Beyer figures is conspicuous given their adoption of TFUS. All due respect to the original namesake. By the looks of the poll (and intuition), looks like you could cobble together a bet-against figure from Beyers and BRIS if you had a contrarian opinion.

I have a conflicted relationship with speed figures. They're the foundation of assessing performance. But they're subjective, vulnerable to error, and they get bet. I try to use them with a wide error bar for evaluation and otherwise try to find reasons to bet against figure horses.

[highnote]

Originally Posted by steveb

assuming steveb is me, then modified as in how...may i ask please?

I read all the posts you made a couple of years about making speed figures and then tried to follow along and re-create what you said. I did not understand some of the methods you mentioned about automating the process using iteration.

The main thing that helped me was that you pointed out an error in the way I was making speed figures. I was not accounting for the difference in time that it takes to cover a length as the distance of a race changes. So the value of a length at one distance may be slightly different than at another distance. I had always known that, but Quirin's method that I was using did not account for it. You also showed that it is better to use the difference between a standard time figure and the actual figure to determine the variant, rather than using the difference between the standard time and the actual time. I believe this is what Beyer describes in his book.

To be honest, I don't know if using Beyer's method over Quirin's method would result in more winners because once you start averaging the difference of actual speed figures to standard figures over the course of the race day to make a variant you have already entered a lot of inaccuracy -- especially if the track is changing throughout the day. This is why some figure makers prefer to make projected variants.

So when I say modified steveb, what I mean is that I read what you wrote and then used the parts that made sense to me. The rest of my process I had to figure out on my own.

So basically, I determine the class par figure and the raw race figure. The difference between them is the race variant. So if the class par is 100 and the race figure is 98 then the race variant is 2 slower than par. Then all race variants for the day are then averaged to create a daily variant that is used to adjust all the speed figures for the day's races. The daily race variant is added to the race figure to determine the final time speed figure.

There are no doubt better ways to do this, but since my way is unique to me, it does add some value.

[classhandicapper]

A good case can be made that Saratoga was very fast early in the card on 7/27, slowed down a bit in the middle, and then speeded up again late. If not, then The Vanderbilt (Imperial Hint) and Jim Dandy may both be both faster than they look (at least on Beyer figures).

That's the thing that gets me so frustrated. The maintenance crews are clearly impacting the speed of the racetrack by watering it, allowing to dry from sun and wind, sealing it, opening it etc.. But they are't doing it any predictable or consistent pattern. If you just average all the race, you are going to have some bad variants and mistaken figures. If you are too liberal with breaking races out, so many of them are so difficult to analyze in isolation you are bound to make mistakes that way too.

Of course the 3rd option is not pay much attention to speed figures at all, but then it gets tougher to recognize that a horse like Shancelot was a major contender in a tough graded stakes despite only facing limited ALW winners at MTH in his last.

It just feels like for every horse they help you with there's another where they throw you off.

[highnote]

Originally Posted by classhandicapper

A good case can be made that Saratoga was very fast early in the card on 7/27, slowed down a bit in the middle, and then speeded up again late. ....

.... Of course the 3rd option is not pay much attention to speed figures at all, ...

... It just feels like for every horse they help you with there's another where they throw you off.

Here's a method of making a daily race variant or even a single race variant.

Maybe CJ's method is kind of like this? I seem to recall someone telling me that Thorograph might use something similar to what I am about to describe?

Compile a list of the days races on comparable surfaces and distances. Then, for a given race, find all the horses that ran today in that race and that were within, say, 2 seconds of par per mile for that race class. These are the horses you want to use to make your race variant.

Next, find the past, say, 5 or 6 races for each horse that are within, say 2 seconds per mile of par. Put their speed figures from each race in a list. Now you'll have a pretty good idea of what they are capable of running today. If there are 8 horses in the race and you use 6 past races from each horse you will have 48 past speed figures. Next, determine the past average speed figure or past median speed figure for each horse from its past 6 races. Determine the race variant for each horse by subtracting each horse's average (or median) 6 past races speed figure from the par speed figure for this race. You will now have 8 variants from which to make today's race variant. You can average them, or find the median, or throw out the highest and lowest and use the median, etc. You can stop here and use this single race variant to make a speed figure for today's race.

Or...

You can repeat this method for all the comparable races on today's card. Maybe you have 5 races on the card that are comparable and each race has 8 horses. That will give you 40 single race variants from which to make a daily variant. You could average them or use the median to find the daily variant. You could do a regression on them to see if there is a trend in the change of the variants over the course of a day.

Be creative!

[cj]

Originally Posted by highnote

Here's a method of making a daily race variant or even a single race variant.

Maybe CJ's method is kind of like this? I seem to recall someone telling me that Thorograph might use something similar to what I am about to describe?

Compile a list of the days races on comparable surfaces and distances. Then, for a given race, find all the horses that ran today in that race and that were within, say, 2 seconds of par per mile for that race class. These are the horses you want to use to make your race variant.

Next, find the past, say, 5 or 6 races for each horse that are within, say 2 seconds per mile of par. Put their speed figures from each race in a list. Now you'll have a pretty good idea of what they are capable of running today. If there are 8 horses in the race and you use 6 past races from each horse you will have 48 past speed figures. Next, determine the past average speed figure or past median speed figure for each horse from its past 6 races. Determine the race variant for each horse by subtracting each horse's average (or median) 6 past races speed figure from the par speed figure for this race. You will now have 8 variants from which to make today's race variant. You can average them, or find the median, or throw out the highest and lowest and use the median, etc. You can stop here and use this single race variant to make a speed figure for today's race.

Or...

You can repeat this method for all the comparable races on today's card. Maybe you have 5 races on the card that are comparable and each race has 8 horses. That will give you 40 single race variants from which to make a daily variant. You could average them or use the median to find the daily variant. You could do a regression on them to see if there is a trend in the change of the variants over the course of a day.

Be creative!

This sounds very familiar. I find it is best to make a detailed, one race variant for every race. Then I compare the most similar races and see if they are reasonably close. If they are, combine them. If they aren't, split them.

I'll start with races at the same exact distance, branch to sprints only and routes only, then see if all are similar. I do the same on turf, and at many tracks today that involves an inner and an outer turf course.

Of course I never combine dirt and turf, but the hope is that all the dirt races are combined and all the turf races on the same courses are combined. People would be surprised at how rarely that actually happens. One thing I've learned is not to force races together that simply don't fit. You will make bad figures the vast majority of the time. If two races look like they were run on different tracks, they probably were.

[classhandicapper]

Originally Posted by highnote

Here's a method of making a daily race variant or even a single race variant.

I agree that using multiple horses per race makes sense, but I still find MANY situations (especially among younger lightly raced horses) when it's hard to know what figure the horses ran because there's not much of a record to go on and the horses are improving. When the track is changing speeds during the day, you really have no idea where a race like that fits. Then when those horses come back you are working with less reliable inputs.

[46zilzal]

Originally Posted by mikesal57

Who is Dave Schwartz?

Creator of HSH software and resident genius of pace and statistical analysis of same.

[mikesal57]

Originally Posted by 46zilzal

Creator of HSH software and resident genius of pace and statistical analysis of same.

A "Genius" would be living here instead in the hills of Nevada...LOL

[highnote]

Originally Posted by classhandicapper

I agree that using multiple horses per race makes sense, but I still find MANY situations (especially among younger lightly raced horses) when it's hard to know what figure the horses ran because there's not much of a record to go on and the horses are improving. When the track is changing speeds during the day, you really have no idea where a race like that fits. Then when those horses come back you are working with less reliable inputs.

Agree. When those lightly raced horses run back you can get an idea of what their earlier figures might have been. The industrious speed figure maker could go back and fix the earlier figures. BRIS used to update their figures at the end of a meet. Not sure if they still do.

[PaceAdvantage]

Originally Posted by mikesal57

A "Genius" would be living here instead in the hills of Nevada...LOL

We get it. You don't like Dave.

[mikesal57]

Originally Posted by PaceAdvantage

We get it. You don't like Dave.

I do.......We are good friends now....but I still bust his balls

Your next!!!

[FakeNameChanged]

Originally Posted by mikesal57

A "Genius" would be living here instead in the hills of Nevada...LOL

Hey, that's where they had the last scene shootout on Beverly Hills Cop. I think it was Dave playing the part of Sgt. Taggert on that. Now why don't we have a best movie of Eddie Murphy somewhere?

[46zilzal]

Originally Posted by mikesal57

A "Genius" would be living here instead in the hills of Nevada..

Read the chapter in malcolm Gladwell's great book The Outliers entitle The Trouble with Geniuses to find out why that is not particularly true..

[Dan Montilion]

Originally Posted by mikesal57

A "Genius" would be living here instead in the hills of Nevada...LOL

That is Dave "The Genius" Schwartz guest house.

[bobphilo]

I was surprised to find the Bris figures to be the most used when in my experience they were among the least accurate. I wonder if their greater use is due to the act that they are compatible with the most software. Particularly with the numerous free software offered by Brisnet.