Quote:
Originally Posted by MJC922
It's a weird phenomena imo but seen consistently enough that it's worth pondering. The thing of it is, the more predictive the model the worse third rank does, so much so that third becomes an increasingly negative trait. It's kind of like when you think of what Quirin did with one of his spot play 'systems' i.e. ranked one two in speed figures and then ranked one two in earnings figures, so the value is in being ranked top two in some model, maybe an entirely different type of model altogether, pace, physicality, connections, whatever, but just from a high level it seems as if something better than third is usually where the value seems to be concentrated. I see what you're getting at with regard to price, I mean such and such is ranked third on class and is 50-1, probably at least some value mixed in there even if third isn't really all that 'positive' in general, it's still probably good enough to win 5% of the time at that number and underneath in exotics might be worth looking at.  |
My guess is that if the any selection method has some value it will typically be concentrated at the top.
All the data we are working with (speed figures, pace figures, class ratings etc...) is flawed to some degree. Looking at 2 horses is better than 1 because if it's close, the 2nd ranked horse may actually be the better horse fairly often. But by the time you get to the 3rd ranked horse the figures are rarely flawed by that much.
However, if for example you use top 2 in speed and top 2 in class, the class ratings will tend to capture the speed figure errors and vice versa. So by combining factors you get more accuracy.
I know from my own database that when I combine my class/form rating with Beyer figures the combined figure outperforms either rating alone. I've been testing that for close to 10 years now. So it's not theory or trial and error. It's fact.
To me the most impressive thing about that is that the class/form rating is computer generated and not nearly as accurate as I could do with a more subjective analysis of field strength. So combining a speed figure with that subjective class analysis would crush just speed figures alone if done well.
What I haven't tested is just top 2 vs. top 3.
