vineyridge wrote:Bryon, I can see a rather large hole.
Horses start their careers and begin to move through them as two year olds. For many of them, their whole subsequent history is determined by two year old results. This has been true for as long as racing has focused on futurities (2 year olds) and the three year old year. Many North American horses don't race after their two or three year old years, and this is something that has become much more common recently. If your model is less predictive for two year olds and younger, it would appear to me to lose a good bit of its usefulness. Wouldn't it be very, very rare for an owner to wait until the 3 year old year to confirm that his/her horse was worthy of training for the very top classes of race? And given how much of the Kentucky Derby money qualifier comes from racing in the 2 year old year, wouldn't that mean that some of your Elite horses don't get entries?
Our Elite horses RARELY race past three.
Not saying that what you are doing isn't very interesting and useful, but if it were more accurate for babies to two, it would be lot more interesting and useful.
Vineyridge,
I am a little confused by some of the terms you have used and I think that you might have misunderstood what is being said. Firstly, while expression may alter as it interacts with environment, the genes themselves don't change from the time that they are born until the time that they die so we can test a foal as soon as it is born and see where, at least in terms of genetics, that it lies in the model. Your statement that the 'model is less predictive for two year olds and younger' leads me to believe that you may think that the DNA changes somehow. It doesn't. Correct me if I have misrepresented you here.
Secondly, the reason why any prediction model is not going to be able to reliably predict early two year old racing is that the competition that the horses generally face at that time is very weak. We are talking about races like the Cinderella, Churchill Downs Debutante, Tremont, etc. These types of races are sometime won by horses that we predict as potentially elite, but often won by horses that we don't predict as potentially elite and these horses that win these types of races usually fall of the radar pretty quickly.
As an example, we predicted Necessary Evil as a
sub-elite horse, she scored right in the middle of the model in terms of class but as a sprinter. She didn't test out as elite on our model, and I don't think that we got her wrong. If she was going through a sale, we certainly wouldn't have put a purchaser off buying her, as she was well out of the non-elite category, but she wouldn't be one that we would go out and recommend. Posse on the other hand, scored out as an elite on the model (he is in the top 5% of the model in terms of class actually). He was a good two year old that trained on at three to compete consistently at graded stakes level which is what our model is trying to identify.
If you are going to create a prediction model, you need to focus the model on what you are trying to predict. We took the view that most people are interested in buying horses that race late in their two year old year at consistent graded stakes level and compete at three and hopefully beyond at the highest level. We created the model to identify this type of horse and it does this very well. We didn't create the model to find horses like Necessary Evil that are out early in their career and for one reason or another, never heard from again. For some people that is a good horse, but for our clients that is not the type of horse that they would like us to identify for them.
