Sidebar

[Dave Schwartz]

Originally Posted by Whosonfirst

Dave, Are you saying that horses within 1 length of the lead at the first call is not predictive of horses who will be 1 length of the lead at 1 furlong from the wire, and hence the win? Or just that it may or may not be an important subset of data? Interesting subject.

The 37% off anniversary sale will be extended through the end of the month.


No, sorry if I was confusing.

Here's what I did:

1. I created 4 models, all based upon the lengths behind at the 1st call.

Specifically...
A. Within 1 length (inclusive)
B. Within 3 lengths (inclusive)
C. Outside of 3 lengths
D. Outside of 6 lengths.

Yes, I realize that A was a subset of B and C was a subset of C. But I was using the factors that were readily available in my system.

2. I queried the system of each subset, asking, "Which factors produce the most winners in the #1 rank?"

IOW, "Of the horses on the lead at the 1st call, which factors best predict that the horse will win?"

Note I also tried other approaches such as top 2, top 3, etc.

The answers I got were pretty astonishing.

For example, only rarely did factors that we'd normally call "important" get into the top 3 of "best factors."

3. As you can expect, the best overall subset to begin with was Subset "A" - those horses within 1 length at the 1st call. Quirin spoke strongly of this in his 1979 book, Winning at the Races.

He basically said that if you could predict who would be on the lead at the 1st call, you would be profitable.

Well, I continued to slice-and-dice.

BTW, I tried the more conventional positional approach... creating subsets from "at which call did the horse 1st challenge for the lead?"

The results were extremely inferior. One problem there was that there had to be 5 subsets: one for each call and one for "Never challenged." Of course, the "never challenged" group were zero-for-everything because it is difficult to win if you never come within 1 length of the lead.

4. The really interesting part was when I built data SEGMENTS above the A-B-C-D. Huh?

The data was segmented based upon the amount of pace pressure in the race, which was a simple approach which I outlined in my Understanding Early Speed seminar.

Basically, there are 4 SEGMENTS - 0,1,2,3+ - representing the number of front-running horses in the race, based upon Early Speed Points.

5. So, it looks like this (still looking at lengths behind at the 1st call)
SEGMENT 1: 3+ ES horses

A. Within 1 length (inclusive)
B. Within 3 lengths (inclusive)
C. Outside of 3 lengths
D. Outside of 6 lengths.

SEGMENT 2: 2 ES horses

A. Within 1 length (inclusive)
B. Within 3 lengths (inclusive)
C. Outside of 3 lengths
D. Outside of 6 lengths.

SEGMENT 3: 1 ES horse

A. Within 1 length (inclusive)
B. Within 3 lengths (inclusive)
C. Outside of 3 lengths
D. Outside of 6 lengths.

SEGMENT 4: 0 ES horses

A. Within 1 length (inclusive)
B. Within 3 lengths (inclusive)
C. Outside of 3 lengths
D. Outside of 6 lengths.

6. In the end we wind up with 16 scenarios, each with a different set of factors.

Perhaps a better way to say it is:

"In a race with 3+ Front Runners...
horses that figure to be within 1 len at 1st call...
win because they have lots of ???"

Same for the other calls.

7. Our HSH users just do this dynamically. That is:

1. They create the SEGMENT by fetching the appropriate races from the database to match the ES Pressure.
(Special note: We use a slightly different SEGMENTING approach because we've got the computing horsepower to do it.)

2. The system is queried based upon each of the 4 possible 1st call positions. (i.e. subsets)

3. The top 3 factors are determined for each subset and a handicapping object is created.

All of this is done with the click of a single button. The entire process takes about a minute.

Hope this helps.

Dave

PS: The HSH software is currently $347 in our store for the rest of April.