Whales: A Little Back Story (blog post)

[Dave Schwartz]

Originally Posted by ranchwest

I think field strength and class are two of the easiest things to program.

Boy, do I agree with this!


In my book, Percentages & Probabilities 2012 I devoted a whole chapter to how to do it with SPEED ratings. (starts on page 57)

Introducing Strength of Field Ratings
Suppose you want to measure the strength of the field in terms of some factor. What the factor is really doesn't matter. It could be a speed rating, a class rating, early speed points, anything. The most logical approach is to simply sum the ratings and divide by the number of horses to produce an average. I'm sure there are many players who do this now.

Personally, I do not like this approach because it assumes that all horses in the field have an equal impact on the outcome of the race. In my opinion, it should not. The problem is that
when you scratch the worst horse in the field, the field appears to get stronger.

Strength of Field factors use an average of the best 4 horses in the race.

[Illustration not shown]

Biggest advantage to using strength over average occurs when the worst horse in the field scratches. If the 9-horse, Queenofthetinkers], scratches the average of the field goes up by over two points to 68.7. In my opinion, scratching the worst horse in the field will have no real impact on the race. In that situation, the strength rating does not change. The number of horses used to produce the strength rating goes down as the field gets smaller.

7+ horses top 4
6 horses top 3
4 or 5 horses top 2

BTW... Percentages & Probabilities 2022 will be available before Christmas.

[ranchwest]

Good points, Dave, especially the evaluation based on the top 2-4 (or maybe 5 in a larger field). I do think, though, that median is often better than average. Sometimes a horse is in a race with figures outside the norm for that type of race. I don't want that horse skewing my evaluation. In my opinion, median gives a less biased glimpse.

[Dave Schwartz]

Originally Posted by ranchwest

Good points, Dave, especially the evaluation based on the top 4 (or maybe 5 in a larger field). I do think, though, that median is often better than average. Sometimes a horse is in a race with figures outside the norm for that type of race. I don't want that horse skewing my evaluation. In my opinion, median gives a less biased glimpse.

There's actually another approach I really like - the 90% Confidence Interval approach that I shared in NewPace.

That came from this book:
How to Measure Anything

Basically, the formula looks at the last 10 races for each horse (or however many are there) and computes likely lowest and highest speed ratings. Allegedly, there is a 90% or higher chance that the horse will run within these ranges.

It's a very easy formula - with literally no math!

But there are problems in the concept as applied because the assumption is that the past 10 races have an equal chance of being run back to. Nothing could be farther from the truth.

However, with some research on which past races are more likely to be repeated, I think it can be made to work.

[ranchwest]

Originally Posted by Dave Schwartz

There's actually another approach I really like - the 90% Confidence Interval approach that I shared in NewPace.

That came from this book:
How to Measure Anything

Basically, the formula looks at the last 10 races for each horse (or however many are there) and computes likely lowest and highest speed ratings. Allegedly, there is a 90% or higher chance that the horse will run within these ranges.

It's a very easy formula - with literally no math!

But there are problems in the concept as applied because the assumption is that the past 10 races have an equal chance of being run back to. Nothing could be farther from the truth.

However, with some research on which past races are more likely to be repeated, I think it can be made to work.

Interesting. I think younger horses might continue to rise, but overall it is an interesting concept.

On which past races might be repeated, I find there is a lot of merit in the Sartin style approach to selecting pace lines... even if one is not using pace factors. A dirt sprinter is more likely, generally, to run back to a dirt sprint race than a turf route, for example. There's other considerations, of course, too.

[classhandicapper]

Originally Posted by ranchwest

I think field strength and class are two of the easiest things to program.

I've never seen any automatic rating I would say is good enough to add betting value.

The entire concept of a "race class" is silly to me and I'm a class oriented handicapper saying that.

Every race has a different strength and depth.

Imagine several Grade 3 stakes.

Race 1. Contains couple of horses competitive at the Grade 1 level, a couple of Grade 3 winners, and a few horses that been competitive in Grade 3 races or won listed stakes.

Race 2. Contains an elite Grade 1 horse using the race as a prep, and several horses that are competitive at the Grade 3 level or recently won listed stakes.

Race 3. Contains a high level Grade 2 horse and a bunch of ALW horses that haven't been competitive in stakes.

Race 4. Contains 8 horses that all look fairly similar having recently won listed stakes or been competitive in Grade 2 and 3 stakes.

Race 5. Contains 1-2 very good Grade 3 horses and a bunch of off form ALW horses.

Race 6. Contains horses that have been getting beaten up in listed stakes and Grade 3 stakes

I could go on and on. The combinations are almost endless.

They all say they are a Grade 3 race, but they can be way different in quality and depth from top to bottom.

Some are very strong top to bottom.

Some are strong at the very top, but average through the middle and bottom.

Some are very strong at the top, but weak through the middle and bottom.

Some are average top to bottom.

Some are of average at the top, but they are extremely deep and have way more contenders than usual.

Some are very weak top to bottom.

You can't easily code for all that when form varies, trips impact results, and the average is almost meaningless in describing the field.

All you can do is subjectively look at the quality of the horses in the race, subjectively look at the trips, and subjectively try to understand the quality of each performance relative to the other horses based on their quality and trips.

Then after the race, you can follow how those come back to run, but imo there's not as much value in that post race analysis because everyone is seeing how they run back and catches on quickly.

I've made attempts to code some of this (strength at the top and depth). The metrics can be useful. But there's no way I can code what I think is required to do a more detailed class analysis of a race that might yield betting value.

[classhandicapper]

Ask yourself this question.

If Seattle Slew won a race labeled as ALW, is that an elite Grade 1 race, an ALW race, or the average of Seattle Slew and a bunch of ALW horses.

I would argue it's none of those.

It's an elite Grade 1 horse beating the crap out of ALW horses.

I don't know how whales can code for all the crazy combinations of things you see. These things get very complex. That's why so many people become speed handicappers. It's way easier.

[ranchwest]

I didn't say that race classification was my only measure of class. But I produce a class number that is the strongest number I have. If I can do it, whales can do it.

[mountainman]

Originally Posted by classhandicapper

Ask yourself this question.

If Seattle Slew won a race labeled as ALW, is that an elite Grade 1 race, an ALW race, or the average of Seattle Slew and a bunch of ALW horses.

I would argue it's none of those.

It's an elite Grade 1 horse beating the crap out of ALW horses.

I don't know how whales can code for all the crazy combinations of things you see. These things get very complex. That's why so many people become speed handicappers. It's way easier.

In addition, field "strength" can be a deceptive concept when, for potentially numerous reasons, the better horses don't fire. And that's not even considering how subsequent performance(s) can alter our pre-conceived notions of field quality.

[ranchwest]

Originally Posted by classhandicapper

Ask yourself this question.

If Seattle Slew won a race labeled as ALW, is that an elite Grade 1 race, an ALW race, or the average of Seattle Slew and a bunch of ALW horses.

I would argue it's none of those.

It's an elite Grade 1 horse beating the crap out of ALW horses.

I don't know how whales can code for all the crazy combinations of things you see. These things get very complex. That's why so many people become speed handicappers. It's way easier.

If there's a good programmer relying on good data, then it is WAY easier to program complexity CONSISTENTLY than it is to find it manually. Yes, there are subtleties that computers won't catch. Computers can't watch replays or post parades. But we're talking about whales analyzing thousands of races and trying to make a small profit (over that huge number of races) that is greatly enhanced by rebates.

Most people won't bother on making a small percentage profit because that barely covers our expenses. If you're betting a million bucks, who cares if you only make 10% after rebate? That's still 100 grand. They can churn through until they make the big scores.

[Dave Schwartz]

Originally Posted by classhandicapper

Ask yourself this question.

If Seattle Slew won a race labeled as ALW, is that an elite Grade 1 race, an ALW race, or the average of Seattle Slew and a bunch of ALW horses.

I would argue it's none of those.

It's an elite Grade 1 horse beating the crap out of ALW horses.

I don't know how whales can code for all the crazy combinations of things you see. These things get very complex. That's why so many people become speed handicappers. It's way easier.

I would say that it is a race made up of a

computable class level

based upon the mix of horses in the race, considering each horse's past races and the computable classes of those races.

Then, on top of that, I would say that Slew was, on a fuzzy scale of 1-5...

5=Excellent
4=Good
3=Average
2=Fair
1=poor

...

About a 9.


Would the computable classes be "right?"
Probably not.
But they'd be at least as good as anything but an honest-to-God class expert would be able to accomplish with a ton more effort.

I also think that most people who would call themselves "class experts" are not anywhere near as good as they think they are.

(Not referring to you, Classic. After all, how would I know?)

[ranchwest]

There's also class factors that don't all come from past races. For instance, the computer is not likely to do a very good job of completely reading the conditions of the race. So, one thing I like is to show some preference for high weighted horses -- they match the weight condition.

Dave, you and I have previously discussed that there are MANY class factors.

[ranchwest]

Admittedly, there is data that a keen eye can pick up.

I recall a friend pointing out to me that a maiden had run close to Risen Star and it won.

Or, a horse I saw ran against Yankee Affair when the other horses in the race would never have gotten a scent of such a horse. This one turned out to be an excellent inclusion on a P3 ticket.

Computers are seldom programmed to pick up such important information. But most manual handicappers aren't that thorough, either.

[teddy]

stock-photo-cool-watercolor-whale-in-glasses-790040773.jpg

[classhandicapper]

Originally Posted by mountainman

In addition, field "strength" can be a deceptive concept when, for potentially numerous reasons, the better horses don't fire. And that's not even considering how subsequent performance(s) can alter our pre-conceived notions of field quality.

Form changes are a problem. The other problem I run into is maiden and other races for very lightly raced horses where the quality is less known.

Speed figures help with those kinds of things, but the primary reason I'm digging into races from a class perspective is to try to find horses that are stronger or weaker than their figures suggest.

I think it's very hard to code for some of things that still offer betting value. That's why I still do it. I focus on stakes to reduce the problems and because it's time consuming. There's a limit to what I can do manually.

[Tom]

Originally Posted by Dave Schwartz

There's actually another approach I really like - the 90% Confidence Interval approach that I shared in NewPace.

That came from this book:
How to Measure Anything

Basically, the formula looks at the last 10 races for each horse (or however many are there) and computes likely lowest and highest speed ratings. Allegedly, there is a 90% or higher chance that the horse will run within these ranges.

It's a very easy formula - with literally no math!

But there are problems in the concept as applied because the assumption is that the past 10 races have an equal chance of being run back to. Nothing could be farther from the truth.

However, with some research on which past races are more likely to be repeated, I think it can be made to work.

Dave, you posted somewhere here a few years ago, a simple method where you used I think 2 SR for each horse, depending on how many races the horse had run. I haven't been able to find it again (but I did find some very interesting threads, like How to Improve Your Game).
Does that ring a bell?