Modeling pace dynamics
 

Computer guys,

How do you model pace dynamics (e.g., early speed battles that set up for closers, lone speed situations, etc.) in your algorithms?

I'm using a rank-ordered logit and currently have a composite factor (linear combination) that projects probability of having the early lead (1/2 mile in sprints and 6F point in routes). Based on this, I can tell roughly where each runner will be early in the race. My problem is, how to add factors for what type of pace scenario is emerging. The model clearly shows that being up front early is on average a positive factor (no surprise here), but does this positive effect vary based on whether there is pace pressure or a speed battle? If so, how could I model this using new variables?

Any thoughts would be greatly appreciated. Thanks!

I'll give you an idea of how you can use these numbers, in races where a lot of horses project to be on or near the lead (horses with similar early pace numbers), instead of looking for the best closers, look for horses that figure to be positioned just behind the early runners. A lot of times, it's not actually the horse with the best late kick that wins. It's the horse that gets the first run on the leaders and passes them while they're tiring. The late closers then kind of keep plugging and finish third and fourth.

In races where there aren't that many horses with sharp early speed numbers, look for good value among the horses that do have good early speed because one of those horses will probably win, either wire to wire or off a stalking or pressing trip.

Good stuff, Pandy. Thanks!