|
Subject: Model refinements
It is clear from last night and even before that, that several handicappers have little or no exposure to Excel and how to use it.
That in itself does not disqualify anyone accessing the data base from developing useful information from it. The current data base has close to 4000 races in it and very few of the values for any one category apply to ALL races. There are several existing tools in the data base that will let you parse the data in many different ways. And you will find the stats for one kind of race can, and often will be, drastically different than other types of races.
Lets say you play Belmont at Aqueduct. You can sort by surface, then by each individual distance from 5 furlongs to 12 furlongs and beyond if you wish. Each sort may have as many as 200 entries and many will only have 10 or so races, so you can see, for instance what the range of % Med is for each distance if you sort by that distance on dirt or turf. I know that the NY turf courses have inner and outer tracks and for now, we do not have the data base set up to delineate those different track configurations. Some may ultimately want to sort the data by the number of horses in the race, some by age, some by sex, etc. And of course these last three categories can be sorted by surface and distance as well. Right now, anyone with a little extra time on their hands could review the data and determine a rough idea of relevant data, until eyestrain catches up with you. (ask me how I know this!)
The more specific the info that is desired, the easier it is to get decent info by reviewing the currently available sorts. The tools for doing such a sort might have to be enhanced, like winners by sex or age, fields of specific sizes, younger horses by time of year (3 vs older, etc.)
One caveat to the current data. This time of year and in the horse industry in particular there are a lot of races with only 5-6-7 horses. The data base does not have any 5 horse races or races with more than two FTS or Foreign entries. So the more the data is influenced by 6 and 7 horse fields, the closer the data will apply if you are handicapping races with those few horses in it. When you come up against races with 10-12 horses the percentage of winners in each category are going to go down because of the increased field size, the number of decisions you make on your way to your top contenders, etc.
There are many factors where one could benefit by knowing MINimum or MAXimim values in a set of information, but it would be more useful if we could set a % age of success. So for something like % Med you could say develop a % Med stat that would get the winner 90%, 85%, 80% or other range of values. Obviously, when you do that you eliminate horses that won when their factor criteria fell out of the % age that you are looking for. Even if one decided not to eliminate a horse for being out of the desired range it should give you notice that the rest of the field should be looked at for other win candidates and hopefully open up the tote board to higher priced horses.
In my comments last night about PRX the MIN/MAX info came in handy for 8F routes in that in 28 races no winners had a % med over 69.5, but for races at 8.3 furlongs the cutoff appears to be a little lower at 69.4 but the data had four outliers. All of these outliers were at 70% or greater, like 70.3, 70.6, 70.8 and all came in those $5K and $7.5K non winners 1X or 2X six month variety. But outliers exist at both ends of the spectrum, thus the need to research reasonable brackets for the info.
Dick had a race the other day that he labeled a CHAOS race, and so he did not play it. That race became playable for me (in theory-I did not play that day) once I ran down the limiting value for % med. Admittedly it was a really ugly race with many having 3F fractions in the mid to upper 40 fps. Those horses could use a sun dial to time there way home! Using Dick's screen shots there were only two horses in the race with a low enough % med to win. One of them did and paid $13.20
There are some factors where max info alone can be used as a demoting influence. Like Days since last race. In a bigger PRX model done a couple of years ago, I found only 2-3% of winners had longer than 90 day layoffs for win and overall the stats for other tracks trended to 5-6%. At PRX and other tracks on the same circuit, it is rare to see low level horses with a long and/or consistent work tab. Those horses get raced into shape, which, by the way, makes it hard to choose a line some times. How many races does a particular horse need before they fire the big bullet? The PPs might give you a clue if you look deep enough.
But even using days away as a guide you will almost certainly find different data for dirt sprint returners, dirt route returners and turf sprint and route returning horses or first time Foreign horses. So even this data has to be framed properly to make a decision.
Knowing what a workout tab should look like to get a horse into shape after a layoff is a little on the art side of things, but there are trainers that do it well and along with PL and BPP info can further limit the number of surprise winners. See KOPION on the feature race at SA the day after Christmas.
Anyway, that is all for now. If I didn't say it last night, I probably meant to....
Yeah, right.
|