MS Access as Handicapping Database

[wiretowire68]

Any possibility of showing me a picture of the exel pp imported into your spreadsheet and a copy of the access data. I am just trying to get an idea of how you do it. I have no idea on how to code functions into my google sheet, similar to exel.. after converting the drf formulator pdfs to csv...It just makes a mess? Help would be great if you could provide step by step if you have time.

Very Sincerely

Chuck

[classhandicapper]

Originally Posted by wiretowire68

Any possibility of showing me a picture of the exel pp imported into your spreadsheet and a copy of the access data. I am just trying to get an idea of how you do it. I have no idea on how to code functions into my google sheet, similar to exel.. after converting the drf formulator pdfs to csv...It just makes a mess? Help would be great if you could provide step by step if you have time.

Very Sincerely

Chuck

I never export the entire set of PPs into Excel. I use some of the data to create metrics and reports to supplement the PPs. Then I’m looking at Formulator PPs, Timeform PPs and my own reports when handicapping.

There is some documentation on the Formulator export files at DRF.

https://www1.drf.com/formulator/userguide.pdf

Here is the documentation on the results files.

https://www1.drf.com/misc/textcharts.html

[wiretowire68]

It is very complicated coding or time consuming to create a pp template from the drf. Thats what I am looking for, similar to Raybo back in the day. I keep trying to find someone who knows how to insert this automatically into a spreadsheet

[classhandicapper]

Originally Posted by wiretowire68

It is very complicated coding or time consuming to create a pp template from the drf. Thats what I am looking for, similar to Raybo back in the day. I keep trying to find someone who knows how to insert this automatically into a spreadsheet

It's not difficult at all.

Once you have the Access database, the export literally takes 1 minute. It's the initial design and building of the database and queries that takes time.

If you want to go straight from the DRF data into Excel, I don't think that would be much harder, but there may be issues with labels for the fields. I haven't done it that way.

[classhandicapper]

I may have learned something new.

Remember, right now I am only testing dirt stakes races.

When I started this project I was looking at Beyer Speed Figure, Class Rating, and Running Style. To my surprise, Running Style did not add a lot of predictive value even though the metric itself as a stand alone does a good job of predicting the leader and winners. That suggested to me that most of the fastest classy horses have early speed. So looking for speed it didn't add much.

Then I added a Jockey and Trainer Rating and running style lost even more predictive value.

I just added Consistency and the predictive value of running style went negative.

That was shocking to me.

What the data is suggesting is that if you have a fast, fit, classy, consistent, horse, trained by a top trainer, and ridden by a top jockey, you are better off if he doesn't have a lot of early speed.

I have a theory to explain it, but I can't be sure yet.

I know from prior studies that high quality races tend to have more sharp speed horses than cheaper ones. So maybe since I'm studying high quality races, they are tilted a little towards races with competitive paces. That would make having speed less of an advantage, but it's still kind of shocking.

When I get to the stage where I am breaking it down by expected pace pressure I should learn more.

[classhandicapper]

Originally Posted by Tom

Class, I attached an example spreadsheet for how I use it, for today's 3rd race at Aqueduct.

It uses the BRIS PP, AC3, and SR last race.
I weigh each according to current results and get a Power number.
The constant is simply the regression line formula for the percentages of the power numbers expressed as probability ranging from .01 to .99.

I'm playing around with the odds line tool you gave me earlier in the thread.

I love the way it works. Thank you.

There's one thing in it I don't understand.

In the "Constant" column you multiply the Power Rating by .0245 and then subtract 1.46. You then convert those results to the probability.

Where did you get the values of .0245 and 1.46 from?

[Tom]

My experience is that the more you add, the less you get.
Every race has particular match up that will affect the outcome, and the key is fiquring out what that might be.

Case in point, the Gotham. A pace collapse was more important than other factors, in that race.. In claiming, n2l, how many proven losers go sown at low odds? How many horses beaten double digits last time win big when dropped from Alw50000 to Cl40 n2l?

Be interested in hearing your theory.

[classhandicapper]

Originally Posted by Tom

My experience is that the more you add, the less you get.
Every race has particular match up that will affect the outcome, and the key is fiquring out what that might be.

Case in point, the Gotham. A pace collapse was more important than other factors, in that race.. In claiming, n2l, how many proven losers go sown at low odds? How many horses beaten double digits last time win big when dropped from Alw50000 to Cl40 n2l?

Be interested in hearing your theory.

I agree with all that. That's way past where I am at this stage with this model.

I intend to try to study more specific situations as I move along.

Right now I'm just trying to build a functional spreadsheet that I can convert into a line making tool that I can also back test.

What I want to do eventually is download Formulator data, import it into the appropriate spreadsheet for the race conditions, push a button, get an odds line, and then make any subjective adjustments I think are appropriate after that.

What I have now produced this result.

2,004 stakes races on dirt
730 top winners (36.43%)
$3768 returned (loss of 5.988%)

So it's outperforming the track take by a huge amount on a pretty large sample.

Hopefully (with a capital H), when I break it down by sprints/routes, sprints with a lot of speed/no speed, routes with a lot of speed/no speed etc... I'll get slightly different weights for each category, better results in each category than the overall results above, and have a better understanding of what matters than I do now. Then I'll move onto turf and other things.

I also want to be able ask, "If I just bet the top selections that were calculated to be overlays by the model, did that produce profits?"

What about 2nd choices?

How did the favorites that were big underlays do?

I'll keep posting things as I learn, but it's slow.

[classhandicapper]

These are the weights the current data and regression test spit out.

[headhawg]

As you add more factors you get more error and (usually) a smaller sample size. My philosophy in handicapping or creating software follows something that Einstein said: "Everything should be made as simple as possible, but not simpler."

[classhandicapper]

Originally Posted by headhawg

As you add more factors you get more error and (usually) a smaller sample size. My philosophy in handicapping or creating software follows something that Einstein said: "Everything should be made as simple as possible, but not simpler."

Ha.

I've been adding 1 major handicapping factor at a time so I could see if there was any benefit. Speed Figure, Class, Consistency, Trainer, and Jockey may have some overlap, but they all operate independently enough to add clear predictive value. The value of running style, not so much, at least in this sample of dirt stakes.

The sample size issue can be a problem if I make the conditions too narrow. I'll have to be careful about that. I'm going to post stuff as I make progress.

[headhawg]

I am wishing you much success. I am interested in seeing your results.

[ranchwest]

Originally Posted by classhandicapper

Ha.

I've been adding 1 major handicapping factor at a time so I could see if there was any benefit. Speed Figure, Class, Consistency, Trainer, and Jockey may have some overlap, but they all operate independently enough to add clear predictive value. The value of running style, not so much, at least in this sample of dirt stakes.

The sample size issue can be a problem if I make the conditions too narrow. I'll have to be careful about that. I'm going to post stuff as I make progress.

Generally, as you add more traditional handicapping factors, you end up with the "best" horse, as does "everyone" else. There's too many things that can go wrong with the best horse to accept the price in race after race. Sometimes, yes... every race, no. Pace makes the race and that means that every race is a new puzzle.

[classhandicapper]

Here's a question for the regression experts.

I have Race Rating.

The winner of the race gets that rating.

I'm then looking at the rest of the field and lowering each subsequent horse's rating by some amount based on each's finishing position and beaten lengths.

Is there a way to calculate what the value of each length and finish position should be and how to weigh them?

What I've done previously is a trial and error run against my database until I found the best value for each. Then a trial and error combining them until I found the best weighting for each. That's probably OK, but I'm wondering if there is a way to ask Excel to simply calculate the values and weighting.

[formula_2002]

Would anyone care to post their $1 returns for your best results.
Dont give away any secretes. Just the following for out of sample races .
# races , # wins and $1 win play return, and or exacta play return.

thanks.