PCR with TSN Race Ratings?

[ranchwest]

If using TSN Race Ratings for computing Scott-style PCR, what would you say is the equivalent of a class step in the sense Scott used?

Does 1.4 RR points seem like a reasonable equivalent of a class step?

[Jeff P]

I would say try it... create a data sample using it... and then make a careful analysis of your results.

Initially I came up with my own point scale just like Scott (Bill Finley Sr.) did in the book.

I started with Bris/TSN class descriptor codes to arrive at race type:

M = maiden claiming
S = maiden special weight
C = claiming
CO = claiming optional
A = allowance
AO = allowance optional
N = listed stakes
G = graded stakes
R,T = starter/handicap

And then within each race type came up with different levels. Example, I used claiming price within M and C race types to award or subtract points. I used conditions such as NW2L and NW3L to do the same thing within allowance conditions.

However, when I rolled everything together and tested PCR in a database I found that win rate and flat win bet roi didn't lead to the kind of profits suggested by the 200 race sample that was the basis of the book.

So I began experimenting.

I found that using a par for the level (available in Bris Files but not TSN) and purse value (despite the fact that people will tell you it's useless to even look at that because purses are fueled by slots) from prior running lines both yielded better results than a point system as suggested in the book.

I'll come back to my take on the reasons for that at the end of this post.

You might also try taking current running line Bris or TSN speed fig earned and using finish position and beaten lengths to arrive at speed fig earned by the race winner from that running line... Race Strength or Strength of Race.

That works well for PCR too. In fact Race Strength is probably easier to arrive at for the class part of PCR than what I use. That said, I opted not to use Race Strength as part of my own PCR calc because I create other Race Strength based variables separately.

I remember that "wow" factor I felt when I first read the book. I used to spend an hour or so per race card manually cranking out PCRs on a TI calculator... and writing the PCR next to the name of each horse in the DRF in bright green ink. That was sometime in the mid 80's if memory serves.

I remember chomping at the bit for the day to come when I could finally get my PCR calc automated. That feeling of getting it automated is part of what drove me down the path that eventually lead to JCapper.

I also remember the disappointment I felt once I finally subjected PCR to hard database testing and began looking at results.

At the time I didn't understand why. At one point I asked myself if maybe the premise behind the book was flawed. After all everything in the book (well written as it was - and I THOROUGHLY enjoyed reading it not once but dozens of times) was based on a sample of just 200 races.

Today I think I have a better handle on why PCR isn't a magic bullet.

My own thought process in this area goes something like this...

Quite a bit of colinearity exists between individual factors used in a typical PCR calc and other factors that bettors are very much in tune with.

For example, horses with lots of 1's and 2's in their positional calls...

Especially in running lines at higher class levels than today's race...

Horses like that tend to be "pretty on paper."

Those same horses often have high beyers... or high other speed figures...

And when they do they generally have low Morning Lines too... and attract gobs of late of money.

Now none of that means that something like PCR has no place in a handicapping model.

Obviously it does.

But like a lot of factors PCR just isn't a magic bullet.

-jp

.

[ranchwest]

I should have mentioned that I was referring to TSN ProCaps files, which do have pars for Race Ratings.

I know PCR's aren't a panacea. I'm looking at various factors that suggest when not to bet high PCR horses. However, I do think PCR can be used as a portion of the solution to the puzzle.

So, I'm still seeking suggestions as to how many Race Rating points should represent one class level for PCR purposes.

I've found that Scott's work is quite useful if you are willing to extend your thought beyond Scott's suggestions.

[Jeff P]

I've found that Scott's work is quite useful if you are willing to extend your thought beyond Scott's suggestions.Absolutely agree with that. Sometimes it pays to be creative.


Does 1.4 RR points seem like a reasonable equivalent of a class step?That SEEMS reasonable. But how can you really know ahead of time?

Answer: You can't.

What I'm suggesting is this:

Create a PCR using your best estimate... in this case 1.4 TSN RR.

And run a data sample using PCR based on 1.4 points.

And make a careful analysis of your results.

But don't stop there.

Create another sample using PCR based on 1.5 points.

And analyze that. Did you improve things or cause them to tank (if only just slightly.)

And generate another sample using 1.3 points. And analyze that.

Let the data tell you how many points to use. After a few samples it usually will.


-jp

.

[ranchwest]

Yeah, I am mostly looking for a starting point because it may be a long time before I have the time to test and refine it.

As a former government employee, the fact that you think 1.4 SEEMS reasonable makes me think it is probably close enough for government work. ;)

After all, I am assuming that PCR has its shortcomings anyway. Plus, who is to say that the Race Ratings are optimal, either?

For anyone else following this thread, note that 1.4 RR points means that the maximum range to be considered for PCR purposes would be +- (1.4x3), for a range of 8.4 points, 4.2 up and 4.2 down.

Thanks for your input.