Modeling Question
 

Good afternoon;

I was looking for some ideas related to modeling. I am currently building models for Saratoga and Delmar. I have a spreadsheet and am working on setting up something to generate an odds line using the strongest factors. It is a points based system. What I'm interested in is to somehow build elimination factors into it. For example, I have a factor for dirt races where currently 98% of the winners are in the top 4 for that factor. My thought was to give negative points to the horses that fall below the ranking of four which would move the horses odds higher. Has anyone tried something like this or have an idea how I could otherwise do this? Your help would be greatly appreciated.

Bill

Hi Bill. I don't think I ever heard of that use for a model. As one who keeps models by surface and dist I stick to the traditional Brohamer approach. I model the 7 primary factors
plus %med,VDC, and 3fr. I note the factor[s] where the winner has the most top 2 ranks. I have found by focusing on top 2 rankings gives a clearer picture of most potent factor[s]. I do same for the place horse. Good luck with your approach.
Tim

Bill, a long time ago, RichieP and I experimented with this type of oddsline-building exercise (assigning negative points), but not sufficiently to know whether it was any better than simply positive points, or top 2 ranks like Tim suggests.

However, it would be extremely easy to experiment with if you have lots of data in the existing standard RDSS Model template. You could start with one of the high-level oddslines (e.g. BLBL or Rx3) then take your factor and assign some negative weight for ranks > 4). The result of the 2 is then a new factor/oddsline, hopefully an improvement over the raw Rx3 for certain race/distance types. Test the hit rate and ROI% over a large enough sample to see if there's an improvement (if any) over Rx3 at each Top 4 tier level.

However, if your factor shows 98% winners on dirt in Top 4, what other factor do you need? What could you do with horses ranked worse than 4th in this factor that would have any useful effect on any other factor (i.e. by assigning them negative points)? Simply find a way to eliminate a 3-1 horse from this factor's Top 4 (when you can), then try dutching the lowest and highest odds of the remaining 3 (or betting the lower odds to breakeven and the higher odds to make your profit).

For Rx3, remember, the number shown is a probability % summing to 100%, so assign negative points accordingly.

If you have a reasonable command of formula building in Excel, you could tinker pretty easily with these ideas.

Good luck!

Ted

Thanks for your input. Here is a little clarification of what I am doing. I'm making a tool to help me in the contender selection process; I'm a vertical horse player, pick 3, 4 and 5's. My current testing is taking the top 3 factors from my win model and plugging the information into a spreadsheet. I've done just what you said Ted and assigned a negative number to contenders that have a rank of 5 and above. My thought is to reward the horses that rank high in these factors and penalize the horses that don't. So far, in my limited testing. it is proving very effective at weeding out the non contenders. I like to play the New York circuit; they have a dollar minimum for a pick three. The easiest thing to do would be to take the top 4 in the top ranked factor and play them; however, this gets expensive quick at 64.00 a play. What I want to do is use the tool to help whittle the contenders down, especially in contentious races. Right now it is very tedious as I have to automate it to get it to work faster. I'll post a screenshot of it later today.

Models
 

I have tried various methods...have figured the percentages of all the horses
and a summation of that. Have tried a certain threshold if they hit above it like 75%. I think what one has to keep in mind with developing models.
Looking backwards at how the model worked is always great...but when you
try it to future races well it all goes to hell in a hurry. This pretty much applies to any statistical method regression analysis,neurotnet/multiple
discriminate analysis it works great for the time period it was built in...but
when you apply outside the time period things do not go so well as planned.
You can look at various factors in the model as Ted and Mitch have mentioned like checking out the CR BL\BL LS V\DL LS
BASED ON 97 RACES a combination of everything the whole kitchen sink.
I came up with CR was like 57% bl\bl=75 v\dl was 77 percent.
that was based on top 4 ...and all at GP... Obviously sample size her is small
it may vary based on surface class distance and even a track. So all I am trying to say is if you develop a model do not depend solely on that model.
Everything can change in a hurry.
Good luck.
And if you do find the Holy Grail in a model please keep me in mind.

You would be better off forgetting about models and find the best corollary verses odds for those top 4 by collecting their ranking and difference in ranking from the best. According to your post, you found one with V/DC. Your sample of 97 races is plenty enough.

With today's RDSS a model is no longer necessary, and that's from Sartin himself. Read my thread " Beginners and Refresher Course." for more tips and from Sartin also. Link below.

http://www.paceandcap.com/forums/showthread.php?t=12316

Analysis of different screens will aide in separating those top 4.

Mitch44

In working with an odds line, I prefer to not use negative numbers. I always get messed up... maybe it's just me. I always add more to the other side of the conditional..

Thanks
 

Thanks for the feedback Mitch.
Always appreciate your feed back...and thanks
for the feedback on other post/threads.
Don

As one who uses models this is what I look for. I look at the primary factors and vdc +%med. I look for the factors where the winner appears the top2 the most. That's where I base my bets on along with % med and form factors. It does not involve %. It works well for me.
Tim

Thanks, dadams6357 greatly appreciated.

Mitch44