On a previous post I mentioned a book I read a couple of weeks ago called "Digital Marketing in an AI World" and promised a couple of thoughts on AI or big data models for horse racing.
If you're interested I'll share a few of those thoughts below.
First, I'm a believer in modeling for sports betting and racing, because with proper discipline, some smarts, and a malleable model we can make hay when the sun shines. Numbers remove bias, and they take away what we thought we knew, but really never knew out of a wager. A good model, even in horse racing's high rake environment, can work, and work pretty well.
When we analyze data in horse racing a few characteristics generally occur.
i) We learn pretty quickly how difficult the game is to beat. The angles we thought were great, aren't great at all (unless you like losing 15 or 20 cents of every dollar you bet). When you run a model on a subset of (statistically significant) data, the chances of it showing +EV are slim. When you parse or layer too much, you're chasing your tail with bad information.
ii) The horses a model may signal as plays are pure overlays, and some of these horses - on paper - will look gawd awful. We'll see an 0 for 11 horse, a bad jockey switch, terrible form - what we may call 'qualitative anti-angles' - that make us not want to wager on the animal (which is precisely why these wagers can approach 1.00 ROI).
iii) the price of the horse (or the bet type) is everything.
The teams we read so much about are doing much of the above religiously. Can we beat them? In my view, Fred's book tells us one way how.
Self driving cars are in the news, sometimes for terrible reasons. When an accident occurs, it sheds light on the problems with AI, and in-turn, some of racing's computer modeling.
This AI works on a three-step process - perceive, plan and execute. In Tempe, Arizona last year, this was on display, and not in a good way.
A woman was walking her bicycle across a four lane highway after dark and was struck and killed by a self driving car. The car's LIDAR perceived a human with a bicycle, but in the planning stage it computed it could not be a human with a bicycle because it's a 4 lane highway at night. In the execution stage the AI asked "what should we do?" and the answer was, "keep going."
Fred writes: "Machine learning systems can make mistakes, and it's possible to outflank the competition by capitalizing on them."
In racing, in my view, we see this often.
Several years ago I remember playing the Woodbine polytrack for the first day of the meet. Like other computer modelers I knew the poly is fast, and horses who make the lead are great bets. At times, even with bad trainers or riders, you could make a score. But, at least one modeler was slow to the draw.
At Woodbine, the fifth race had a horse who my model said would make the lead, and horses who made the lead were 4 for 4* already. It was a green light. But one bot, run on a model, didn't agree. Whether it was working with late pace numbers, didn't have a built-in bias, or what I do not know, but it just kept fading the animal. It was 6-1 on the board 14-1 on the exchange, then 16-1, then 17-1. I kept putting up cash, wondering what in the hell price this model was working on. It took the offers up to 25-1. The horse won easily and paid $18.
The above is not as isolated occurrence.
What other mistakes do we see?
In my view - lame horses. If you're an old school horse watcher, you can take these models to school, using the model itself - perceive, plan, execute. Did the horse look like this last time? No, then execute, because that model betting $1,400 on its nose likely has no clue.
I have seen horses head to the gate lame who were gate scratched that the models were on. They're losing money, so we have to be the one to beat them.
Overall, I am certainly no expert, but I love modeling and models, for horse racing or otherwise. They work. But, it doesn't mean they can't be exploited. Those are a couple of areas I think they can be had.
Have a really nice Friday everyone.
* My top pace figure (with a slight speed track modification) went 10 for 10 that day. It was a rare Let It Ride type day that keeps a lot of us coming back.
Sinking marketing money directly into the horseplayer by seeding pools is effective, in both theory and practice In Ontario and elsewher...
One of life's many mysteries on gambling twitter is the Jackpot Bet. Oftentimes people like @shottakingtime, echoed by others, will pos...
Our wagering game is an incredible mental exercise for many reasons. And one of its characteristics I like best is the variety of thought wh...
There's something going on in horse racing today , but I have not really followed it. Instead, I've been thinking about two words we...
At Northlands Park Wednesday several rabbits were in the infield. This spooked one of the horses, and he took a tumble, sending driver Debbi...
Unless you are off the twitter grid (God bless you), you've no doubt witnessed the feud of the month(s) between ITP and some public raci...