Track Profile Handicapping

[Saratoga]

Say you have 10 non-correlated factors.......

Also say you have the last 3 meets of data......

How would you profile a track?

Lets use Woodbine , for example.....

How do you proceed for the up-coming meet ?

This is what I see as choices:

1) Somehow get all 3 meets combined....

2) Use last 2 combined

3) Use last meet

4) Wait a month or so on new meet then compile...


My thoughts , with 1-2-3 your going to get some sort of a "middle" of the

road average and hope that the track plays that way on that day your betting

With #4 , it a "what's working now" scenario that you might get.....

Some more issues to think about...

Do you break it down by distance ,surface , and or class

Why? , I belive that maidens are a different ballgame compared to winners

Thoughts?

[Saratoga]

Still thinking ?

I tried a few scenarios I mentioned but I come up short on them...

[headhawg]

Sample size is king in any research. In theory, using three meets of data is going to be more reliable than one. I understand your concern about recency and possible trends. Ideally you would want to give more weight to the recent meet over older data. I have no idea how to do that or if it will render the analysis useless because the raw data has been manipulated. A stats expert would know.

In short, unless the track surface has changed significantly I would go with more data rather than less. However, there are probably small sample statistical analysis strategies that would still give reliable results if you want to do breakdowns by class, surface distance, etc. Again, consult a stats expert.

[Tom]

Find out what fctors are important.
Contact Dave Schwartz about his Find Out What's Winng idea. Best thing I've ever seen for this project. Kinf like a Brohaner model on steroids.

Set up your data in Excel or equivalent so you can use filters.

Suggested Col headers (rank each winner)

Date
Distane
Surface
Class
Sex
Age
Rank, Factor 1
Rank, Factor 2 etc
Rank, Factoe 10
Payoff
No. horses

Now you use Dave's formula to rate each factor.

I looke at dist, surf, class, age.

I'd do each year seperately to see how consistent over time a factor is.

You can make a power rating for say your top 4 factors:

Factor7 45%
Factor3 25%
Factoq8 20%
Factor2 10%

[Saratoga]

Originally Posted by headhawg

Sample size is king in any research. In theory, using three meets of data is going to be more reliable than one. I understand your concern about recency and possible trends. Ideally you would want to give more weight to the recent meet over older data. I have no idea how to do that or if it will render the analysis useless because the raw data has been manipulated. A stats expert would know.

In short, unless the track surface has changed significantly I would go with more data rather than less. However, there are probably small sample statistical analysis strategies that would still give reliable results if you want to do breakdowns by class, surface distance, etc. Again, consult a stats expert.

Originally Posted by Tom

Find out what fctors are important.
Contact Dave Schwartz about his Find Out What's Winng idea. Best thing I've ever seen for this project. Kinf like a Brohaner model on steroids.

Set up your data in Excel or equivalent so you can use filters.

Suggested Col headers (rank each winner)

Date
Distane
Surface
Class
Sex
Age
Rank, Factor 1
Rank, Factor 2 etc
Rank, Factoe 10
Payoff
No. horses

Now you use Dave's formula to rate each factor.

I looke at dist, surf, class, age.

I'd do each year seperately to see how consistent over time a factor is.

You can make a power rating for say your top 4 factors:

Factor7 45%
Factor3 25%
Factoq8 20%
Factor2 10%

Thanks Brian and Tom....

You both hit it on the head as I have tried and worked with those exact approaches .........

I did see the non-consistency issue over the years with the factors

which makes it that much difficult , which is why I tried the "recent" road...

Seems like I'm banging my head most of the time...

With Daves formula Tom, how much different is it than using Excel's Regression or Solver functions...

Hope the stats guys chimes in...

[headhawg]

Off the top of my head...why not three different analyses? Use logistic regression to get your factors/weights based on one year, two years, and three years of data. See how many factors each method has in common. In a perfect handicapping world you would have little or not change in the factors between each analysis. If there's a significant difference then the most recent might be the most reliable. Going forward just keep track of what's winning, either using the method Tom is suggesting or by your betting records -- are you winning or losing?

[Dave Schwartz]

Originally Posted by headhawg

FROM HEADHAWG:
Off the top of my head...why not three different analyses? Use logistic regression to get your factors/weights based on one year, two years, and three years of data. See how many factors each method has in common. In a perfect handicapping world you would have little or not change in the factors between each analysis. If there's a significant difference then the most recent might be the most reliable. Going forward just keep track of what's winning, either using the method Tom is suggesting or by your betting records -- are you winning or losing?

^^^^^^
THIS.

FROM TOM:
I looke at dist, surf, class, age.

I'd do each year seperately to see how consistent over time a factor is.

You can make a power rating for say your top 4 factors:

Factor7 45%
Factor3 25%
Factoq8 20%
Factor2 10%

^^^^^^
AND THIS.

To add to Tom's post idea...

We humans typically find it difficult to think in terms of percents because there are so many possibilities.

Think of them as Fibonacci Multipliers.

1.00
0.62
0.38
0.24
0.14
0.10

(You can Google Fibonacci if you need to.)

So, now you have up to 6 factors...

Instead of choosing 6 weights, you ask 6 very simple questions:

1. Which factor is MOST IMPORTANT?
2. Which factor is 2nd MOST IMPORTANT?

etc., down the line.

This also addresses (to some degree) interdependency between the factors.

Not saying it's right.
In fact, in a given race, I can all but guarantee that the weights are wrong. But they are good enough.

.

[ranchwest]

Model %Median by distance and surface, establishing a preferred range. Only consider horses in that range.

[Saratoga]

Originally Posted by headhawg

Off the top of my head...why not three different analyses? Use logistic regression to get your factors/weights based on one year, two years, and three years of data. See how many factors each method has in common. In a perfect handicapping world you would have little or not change in the factors between each analysis. If there's a significant difference then the most recent might be the most reliable. Going forward just keep track of what's winning, either using the method Tom is suggesting or by your betting records -- are you winning or losing?

What I did:

Got 3 years in a database...

Pulled up top 20 factors using the ...ROI x Win% approach...

Ran it Excel"s Regression....It showed my factors in order eliminating the correlated ones...

You know guys...I'm going to continue this using BetMix because it can do things ..FAST

[headhawg]

Does BetMix have your 20 factors in the 40 or so factors they use? I didn't look long at the site, just a curiosity glance tbh. Can you customize the factors (create your own) and run them using their data?

[Saratoga]

Lets do the Woodbine Model....

Criteria:
Last 3 meets
3+ year old
Dirt Sprint
Claimers

Picture #1......Number of plays and stats

Picture #2.....Top % Factors in Rank order....

more....

[Saratoga]

First issue....

SPEED category is most plentiful here .....

Can't tell which are correlated....


There are 7 categories to choose

Speed has 11 factors

Pace has 8

Earnings 6

Train/Jock 13

Ped 4

Form 6

Class 4


My question here is to use factors from each group or use top down in picture 2?

[Saratoga]

Lets use TOP % FActor "Last Purse" Rank..can't use OFF Odds cause you don't know it...

I don't like using Morning Line because everyone see's that...am I wrong?

Rank 1

Rank 1-2

Rank 1-2-3

This looks like Rank 1 , but remember the number of plays drops fast when using low ranks..

You just went from 6000+ to 1300 in one shot...

[Saratoga]

Heres first dilemma....

After using LAST PURSE......the next 4 are class based.....

I believe to skip them and go to "Lifetime Earnings" which is next in line.....Correct?

Following me?


Next Lifetime Earning...

1

1-2

1-2-3

#1 looks right again and now we are in positive ROI

and number of plays are now 283.....for 3 years

see where I 'm getting ?

Do I stop at a + ROI or do I need more or do I expand LIfetime Earning to Rank 3 to get more stuff into it?

[Saratoga]

I'll stop now for comments...