The Face Value of Ratings

[MJC922]

Scaling is cosmetic. For example Beyer pinning 10k open claiming back in the day to an 85 and working up and down from there for each fifth was not really meant to convey a percentage 'better'. If you want to think in terms of clock time then any Beyer figure could be converted back into say a 6f time like 1:12-2/5 so then the clock times could be used in your handicapping instead of a speed figure. Nothing particularly wrong with that thought but your Derby winner let's say ran the 10f final time equivalent of 6f in 1:08 is kind of an odd way of looking at it in the PPs, I guess it's just a personal preference.

Again the baseline figure is arbitrary, we could use 185 or even 900 instead of 85 for a 10k open baseline par at all distances as long as the points per fifth are correct at the different distances then it all works out in the end. It works out that is if you're a 100% true believer in proportional time theory, which I no longer am but it generally works ok up to a certain point.

Looking at my class ratings for example, I wanted a horse that reaches 100 to be considered graded stakes caliber and I also wanted the scaling to be kind of roughly close to handicap weights. It's mostly just cosmetics. The worst horses in NA score in the 20s and the best usually reach 120, well Flightline hit a record 131.49, I'm going to make that into a t-shirt btw if anyone's interested. I'm working on it.

[fast4522]

Originally Posted by ranchwest

I understand that it was tough to find appropriate pace lines, but Bris Prime Power had it, my odds line had it, good works, figured to stalk and close. I'd only play the race as a horizontal where I needed to include it, but it is gettable.

What is gettable I stopped doing over 10 years ago, what is value that fits the parameters is my thing. You pump in the last 300 races and I could care less about that data. Many a runner I bet that was over $10 that paid less than $10 are just the breaks. I need to get over $10 to win and I do well. $6 horses are like masturbation, have at it. Sell software that has gettable $6 , $7 , $8 with your data to back it up, I will pass and stay with what I am doing. No one sells focus with handicapping programs, few learn to do it. I have no choice because I like playing the ponies. I answered 46zilzal question.

[ranchwest]

Originally Posted by fast4522

What is gettable I stopped doing over 10 years ago, what is value that fits the parameters is my thing. You pump in the last 300 races and I could care less about that data. Many a runner I bet that was over $10 that paid less than $10 are just the breaks. I need to get over $10 to win and I do well. $6 horses are like masturbation, have at it. Sell software that has gettable $6 , $7 , $8 with your data to back it up, I will pass and stay with what I am doing. No one sells focus with handicapping programs, few learn to do it. I have no choice because I like playing the ponies. I answered 46zilzal question.

As I said, I would only play the race if I needed it to be a part of a horizontal sequence. Your response suggests that you do not play horizontals, which puts us on the same page as to the individual race.

For horizontals, one need only find a few value plays to make the entire sequence have value.

Different strokes.

[classhandicapper]

Great post Jeff. I’ve been grappling with some of these problems for awhile. I put my regression analysis project on hold because I wasn’t coming up with good solutions to some of my problems. This will help stir my thinking again.

[classhandicapper]

Originally Posted by sjk

It would be much harder today as the competition is already there. My return has gone down roughly 1% a year for the past 20 years using exactly the same process.

Are you finding fewer overlays, smaller overlays, or both?

By the way, I want to thank you and others again for all the help you gave me back in 2014 when I was starting my database.

[bisket]

I look at an entire day of races as often as I can to see the data behind the figure or the rating. Equibase historical charts is my best friend. I think some days figures are based on much more solid data. I only play big race days. So figures are more valuable for me on days when the best horses are running in multiple races. I don't have to make a figure for every horse, every day, so I have the freedom to do this. When Beyer invented figures there was much more solid data on a daily basis. The vast majority of races were on dirt so you had much more daily data. Now half the races are on dirt the other half on turf, less data every day. On top of that you have off the turf races moved to dirt. I have the freedom to put a line through them. A figure maker doesn't necessarily have that luxury if that's the only race he has to go on. He's going to handicap horses the entire day to decide on a figure. This becomes a slippery slope when done too frequently. I'm not a daily player so when I choose a race to bet I have much more time to look these type of things than someone who plays every day. I can pick and choose the value in every figure. I think we all have our own criteria for value. That's what makes the mutual pool more like playing chess than any other bet.

[sjk]

Originally Posted by classhandicapper

Are you finding fewer overlays, smaller overlays, or both?

By the way, I want to thank you and others again for all the help you gave me back in 2014 when I was starting my database.

I am finding fewer playable races to the point where it is arguably not worth the time to track so I am at it for fewer hours. The fields are shorter. But mostly the races are bet a whole lot sharper than they used to be.

In the old days there would be lots of large fields and often there were a couple mid-priced overlays and lots of interesting combinations.

These days my plays are simpler and smaller since the expected win percentage is half what was in the old days. The wins are relatively sparse but if all works out the avg price is high enough to get it done.

I took some lumps when GP put in tapeta so I am done with those races and Woodbine too.

I have been pleased to read how you have utilized your database with good success.

[classhandicapper]

Originally Posted by sjk

I have been pleased to read how you have utilized your database with good success.

I learned a lot new things and was able to verify a lot of things with data that I "thought" were true from years of trial and error learning, but like you I'm finding even fewer bets than before. Most of those seem marginal. I'd be better off putting more energy into other things, but I love the challenge. The recent high profile breakdowns have me reevaluating whether that kind of pain makes the pleasures worth it.

[ARAZI91]

The problem here is that horse race ratings and i'm talking of ratings such as those scaled 0-140 or 0-135 like these or Timeform's do not follow the "Normal Distribution" - Ratings such as these which are the WTTR Global World Rankings
WTTR RANKINGS.JPG

The inputs used - Raw times, DTW(Distance To Winner), Field Size , Class,etc all follow different distributions so therefore once the ratings are compiled , they don't follow the Normal Distribution (if you think back to school and the teacher would take the pupils heights and show it as a distribution it usually followed the Bell-Shaped curve of the Norm.Dist. Like this
NormDist.JPG


Horse race Ratings such as the above actually follow the "Gamma Distribution"
Beaten Lengths come in at intervals behind the winner, with the greatest concentration coming just after the winner and the gaps tending to get wider as the back markers trail home. A ten-runner horse race, for instance, isn’t decided by ten individual time trials. If it were, we could describe the finishing times by a normal distribution, such as the one that describes pupils heights in the Classroom again, for instance.

However you can through math and statistics "Brute-Force'' your ratings to something that approximates the Normal Distribution.

One idea is to use a "scoring function" in the form of a ratio which would also derive the max amount a horse can be beaten as well as give you the first length value in a race - here is one over a mile race (circa 100 secs) with the scoring function at the top
typical im race over a 100 secs.JPG


The ratio of x = 45*Beaten Lengths/Beaten Lengths +12 means that your first length value for the winning horse will be 3.75 lengths - it also means that no horse in the race will be beaten by a Max of 45 lengths (over a mile remember - at different distance length values and ratios will change)

So looking at an actual race the ratings would look like this
ACTUAL RACE 10.JPG


Note that the also rans (non winner) would have different length values in a monotonic decreasing fashion as they should be

Here i've calculated the first five home and their respective length values
as you can see they follow the monotonic function of g(f(x) (decrease/increase by interval of distance)
nd Ratings.JPG

[Dave Schwartz]

ARAZI91,

Excellent post.
Your knowledge of statistics is far above mine.

At least I have enough knowledge to visualize the value of your insights.

Lately, I've been seeing the value of confidence intervals using the method from How to Measure Anything

Amazing how well CIs can be used to create new factors that turn out to be highly relevant.

[Bustin Stones]

My approach has drifted away from spd ratings in general. For races under 6f I generally look at sprint raw times. And for 6f and above I ignore all things related to speed. Most of those races look to me like velodrome bicycle races. The horses are going just fast enough to get to the final 2f to set up a sprint to the wire. The overall time or rating becomes misleading. There are far more meaningful clues like class rating and race rating of recent races that tell me whether a horse will be competitive in those final furlongs.
I do understand the attraction to speed ratings in that they afford the chance to massage data. The value of any data is not related to their capacity to be massaged. I get that for Mdns or 2yr olds you have to use spd ratings because there isn't much else to go by.

[classhandicapper]

I'm not so sure I like artificially adjusting the value of a beaten length or position to make it fit a desired distribution.

There are reasons beaten lengths tend to vary by how far back the horse finishes.

1. Well beaten horses that are no longer in a position to compete for better purse money will often be eased late.

2. Horses that are overmatched generally don't distribute their energy efficiently. They get used up early chasing and trying to keep up with superior rivals and wind up running a slower final time than they would have in a race against similar rivals.

It seems to me if you want to adjust for things like that, you have to visually look at a race to see when and how severely a horse was eased, but more importantly work with some kind of pace figure or pace adjustment. Then you aren't making an adjustment that approximates the average. You making an adjustment based on the realty of that individual race.

If you look at TFUS figures in a chart, you can see how the pace formula tries to adjust for the impact of pace component.

The other thing some people do is simply ignore a race where the horse was clearly badly overmatched and use a figure from a prior race where the horse fit well with the competition.

[sjk]

I have had good luck making meaningful statistics out of this sort of thing. Each horse in a race has his own race shape and if you take a large sample of horses with that race shape you can evaluate the degree to which the race shape aided or compromised the runners.

Rather than making an artificial adjustment to the beaten lengths you are assigning a statistic based on inefficiency of energy distribution that you can combine with your speed figures etc in some way.

[classhandicapper]

Originally Posted by sjk

I have had good luck making meaningful statistics out of this sort of thing. Each horse in a race has his own race shape and if you take a large sample of horses with that race shape you can evaluate the degree to which the race shape aided or compromised the runners.

Rather than making an artificial adjustment to the beaten lengths you are assigning a statistic based on inefficiency of energy distribution that you can combine with your speed figures etc in some way.

I've had some luck with that making comparisons between horses within the same race that had a similar easy/tough trip, but I still haven't come up with a good way of making those comparisons across different races other than to say Horse X is better or worse than he looks. That's one of an endless list of things I'd like to study but don't have the time for.