Quote:
Originally Posted by Running Amok
...And I'm also asking if you wanted to assign a number to represent the intrinsic value of a rating, such as a speed rating, or any other type of rating that takes a measure of something and convert it to a number, how should it could be calculated?
In the world of finance, intrinsic value is a way of describing the perceived or true value of an asset. I would like to know the perceived value of the ratings in the PPs.
By the way, I can tell by the replies I've got that there's a lot of really good handicappers in this forum. And guys, I'm just trying to think outside the box here. Just trying to look at things in a different way than the conventional view.
|
I found myself asking these same questions some 30 years ago.
The answer I decided on (your mileage may vary) involves:
1. Getting the raw data into a database.
What data?
I wanted raw data covering many different aspects of the game.
From past running lines: early pace figs, late pace figs, final time speed figs, position at each call, fractional time at each call, beaten lengths at each call, class level, purse, post position, weight carried, surface, distance, runup distance, track condition, and literally hundreds of other data points.
Also workout histories, records of the sire/damssire/dam, rider/trainer, and of course tote data.
From there I began creating my own custom factors from the raw data - and (eventually) started getting my own custom factors into my databases.
2. Transform the data as needed.
What transformations?
Many factors exist on different scales. For example, depending on whose speed figs you are using final time speed figs for most horses might be between 60 and 130.
At the same time velocity of the horse at various points of call might be 58 fps at the 2f point of call, 56 fps at the 4f point of call, and 52.5 fps in the stretch.
Depending on whose pedigree ratings you are using numeric pedigree for each horse can be orders of magnitude higher than speed and pace figs.
Same thing with purses.
At the other end of the scale win percentage for post position, rider, and trainer, etc. for each horse in any given race might range from 0.02 to 0.38 (ballpark.)
Are there statistical techniques you can use to smooth out scaling differences among multiple factors?
Ron Tiller (above) mentioned Z-Scores. I actually use Z-Scores for some of these situations. I also use exponential equations for others.
There's no one right answer to this. Think of it as an engineering problem.
You try something and see how well it works. Try something else and measure how that works. (Compare the two.)
You end up keeping what the data tells you works best, and you end up discarding what the data says doesn't work. (It's a constant iterative process.)
3. Clean the data to handle outliers.
What outliers?
Many races have horses with no speed and pace figs. For example, first time starters, foreign shippers, and even the occasional horse that raced on a day when the chartcaller couldn't see the horses on the other side of the track because of fog, etc.
These horses are outliers because they have 0's for their speed figs, pace figs, and null values for positional calls, and beaten lengths in the data.
But you know they are not likely to earn a 0 speed or pace fig in today's race.
So how do you treat them?
One way, for each factor, is to calculate the mean for the horses that do have numbers - and assign each outlier horse a factor value approximately equal to the mean.
Another way is to simply remove all races from the data that contain one or more outlier horses before performing statistical analysis.
4. Export clean data to .csv file and run regression analysis (linear regression, multiple logistic regression, or conditional regression) to get a basic understanding of the relationship among the variables.
5. Create a model based on the output of the statistical analysis software.
The basic process for using logistic regression to create a horse racing model is covered in the book
Precision by CX Wong.
And no. I don't recommend creating the same factors Wong talks about in his book. (I recommend rolling your own and creating your own factors.)
But I do recommend the book for anyone who is serious about creating their first horse racing model. Imo, reading the book will give you some ideas and help kick start the process.
You can spend a lot of money on statistical analysis software.
You can also use many of the free packages in the r programming language, or even something much simpler like Solver in Excel.
No matter how you do it, the purpose of statistical analysis is to understand the relationship among the variables.
One of the outputs generated by statistical analysis software after max likelihood estimation during logistic regression is a beta coefficient (which can be used as the 'weight' for any given factor.)
Imo, the beta coefficient for a factor is probably about as close to the concept of intrinsic value from the financial world you alluded to.
And of course the beta coefficient for each factor changes as you change the factor mix of the model. (Think relationship among the variables.)
Wrapping Up
If you decide to do this, realize going in, you will likely spend hundreds (if not thousands) of hours - with no guarantee whatsoever of success.
But also realize (Imo) what I just posted is a
general description of the road taken by the largest handle volume horse bettors on the planet.
I would add this is probably not the roadmap to success for most players.
It's a roadmap that will have you butting heads with the smartest bettors on the planet.
In order to truly succeed you'll have to be better than them. (And they are really good at what they do.)
But I did want to post my thoughts on the question that was asked:
Quote:
|
"In the world of finance, intrinsic value is a way of describing the perceived or true value of an asset. I would like to know the perceived value of the ratings in the PPs."
|
And yeah.
My favorite hobby?
Flyfishing.
-jp
.
