The Gambler Who Cracked the Horse-Racing Code

[GMB@BP]

Originally Posted by Dave Schwartz

Yes, lots of thoughts.

First, you're looking at it backwards. The percentage of whale money is much higher at small tracks because the rebate is so much higher.

Sure, they bet less, but they bet a greater percentage of the pool.

Think of it like you are walking into a poker room.

There are two games, both with 6 players and the same rules.

Table A: Lower rake, populated by 3 strong players.

Table B: Higher rake, but no sharks.

I would suggest that B would be an easier table to beat.

Same with racing.

My experience is that my ROI is better at tracks with little or no rebate. An added benefit is that a greater percentage of the time the winner's odds go UP in the winner's circle.


Dave

Do you know what tracks have the worst rebates, in general?

I would imagine lower take would usually mean less rebate but honestly its a pure guess on my part.

[cj]

Originally Posted by GMB@BP

Do you know what tracks have the worst rebates, in general?

I would imagine lower take would usually mean less rebate but honestly its a pure guess on my part.

It has to do with takeout but also signal fees. The best tracks have the highest signal fees so generally the worst rebates.

[classhandicapper]

Originally Posted by cj

Jockey tendencies can certainly be analyzed and exploited. Could you give an example?

They probably have good jockey data, but I'd wonder if they have any data on which jockeys, trainers, and owners pick up on biases quickly and adjust and vice versa.

To get that refined, you'd have to watch a specific circuit closely enough to have good bias notes and be able to observe who seems to be riding aggressively to avoid a negative bias or to take advantage of a favorable one. I don't keep information like that either, but from time time I make mental notes. That's the kind of thing that's probably not useful often enough for them to care, but if you have 10 things like that you might have the beginnings of a portfolio of info with value.

[cj]

Originally Posted by classhandicapper

They probably have good jockey data, but I'd wonder if they have any data on which jockeys, trainers, and owners pick up on biases quickly and adjust and vice versa.

To get that refined, you'd have to watch a specific circuit closely enough to have good bias notes and be able to observe who seems to be riding aggressively to avoid a negative bias or to take advantage of a favorable one. I don't keep information like that either, but from time time I make mental notes. That's the kind of thing that's probably not useful often enough for them to care, but if you have 10 things like that you might have the beginnings of a portfolio of info with value.

I don't doubt there is still stuff that can be found and used. I do it myself.

The kind of stuff you mention could be programmed, at least somewhat. If somebody is good at noting biased tracks, and I do think that can be automated, it isn't too hard to test the rest. Jockeys that are good on biased tracks, not good, even specific types of biases and at specific tracks.

Again, I'm not saying it is being done, but it certainly could be done.

[Maximillion]

Originally Posted by cj

I don't doubt there is still stuff that can be found and used. I do it myself.

The kind of stuff you mention could be programmed, at least somewhat. If somebody is good at noting biased tracks, and I do think that can be automated, it isn't too hard to test the rest. Jockeys that are good on biased tracks, not good, even specific types of biases and at specific tracks.

Again, I'm not saying it is being done, but it certainly could be done.

If you were to automate something like a biased track ,what would be your ballpark guess as to percentage of races or days that would get flagged as such?

[cj]

Originally Posted by Maximillion

If you were to automate something like a biased track ,what would be your ballpark guess as to percentage of races or days that would get flagged as such?

Probably varies by track. First, you'd have to decide what constitutes a bias. Some tracks have long term biases, as in more or less speed favoring than average. It can get really complicated.

For example, track A could really favor speed, track B more fair, so shipping between the two would be one type of bias. Then there are short term biases that crop up everywhere at times. It can get really complicated.

That said, with a good programmer, it could all be done. I would look for days/tracks that show up in the top and bottom 10 or 15% personally.

[GMB@BP]

Originally Posted by cj

Probably varies by track. First, you'd have to decide what constitutes a bias. Some tracks have long term biases, as in more or less speed favoring than average. It can get really complicated.

For example, track A could really favor speed, track B more fair, so shipping between the two would be one type of bias. Then there are short term biases that crop up everywhere at times. It can get really complicated.

That said, with a good programmer, it could all be done. I would look for days/tracks that show up in the top and bottom 10 or 15% personally.

I would imagine they have teams that analyze all this stuff.

One guy reviews trips and they input an adjustment for that,

Another guy reviews workouts and they input for that.

Maybe another guy gets to know the barns and talks to people, they could input for that.

etc etc

[Maximillion]

Originally Posted by cj

Probably varies by track. First, you'd have to decide what constitutes a bias. Some tracks have long term biases, as in more or less speed favoring than average. It can get really complicated.

For example, track A could really favor speed, track B more fair, so shipping between the two would be one type of bias. Then there are short term biases that crop up everywhere at times. It can get really complicated.

That said, with a good programmer, it could all be done. I would look for days/tracks that show up in the top and bottom 10 or 15% personally.

'
Thanks.

[cj]

Originally Posted by GMB@BP

I would imagine they have teams that analyze all this stuff.

One guy reviews trips and they input an adjustment for that,

Another guy reviews workouts and they input for that.

Maybe another guy gets to know the barns and talks to people, they could input for that.

etc etc

That is what I think too. I think many are underestimating what can be done with a team that does nothing but horse racing.

[GMB@BP]

Originally Posted by cj

That is what I think too. I think many are underestimating what can be done with a team that does nothing but horse racing.

Very well capitalized as well, would be very easy to get information with money to spend.

[classhandicapper]

Originally Posted by Maximillion

If you were to automate something like a biased track ,what would be your ballpark guess as to percentage of races or days that would get flagged as such?

I have speed/closer bias automated, but I don't think it can be automated really well.

For example

If the rail is bad, quite a few speed horses will probably wind up on the rail and do poorly.

If it it's a closer track, quite a few speed horses will probably wind up on the rail and do poorly.

The algorithm looks at the result.

Was it a dead rail or a closers track or both.

That's going to take an analysis of how the speed horses off the rail did, how the closers on the rail did, what the paces were, how good the horses were etc..

What about horses that spent some time inside and some time outside?

What about if only on section of the track is bad (like the turn)?

My algorithm spits out degree of bias (1 or 2 standard deviations away from the norm will get flashed), sample size, and confidence level, but I override it plenty of times after I analyze the races.

IMO, it's probably not best to think in terms of there being a bias with some certainty. It's probably better to think in terms of probability because the evidence will sometimes be mixed or limited.

I look at the evidence and instead of just saying it was a bad rail, I might say it's 80% the rail was bad, and 40% it was really bad. Then you could always watch how the horses do next time and adjust.

[dilanesp]

Originally Posted by cj

Jockey tendencies can certainly be analyzed and exploited. Could you give an example?

Forty Niner being sent in the Preakness

[dilanesp]

Originally Posted by classhandicapper

I bet on mule races.

I can't bet a lot of money because the pools are tiny, but I see inefficiencies all the time between the exotic pools and win pool. It's not easy to exploit because the late swings are so huge, but I doubt any whales are in those pools.

No, but I hear mules bet on whale races.

[Dave Schwartz]

Originally Posted by cj

It has to do with takeout but also signal fees. The best tracks have the highest signal fees so generally the worst rebates.

As usual, I second what CJ said.

In fact, most large players' rebates are directly tied to what the ADW would make minus what they get to keep.

classhandicapper
They probably have good jockey data, but I'd wonder if they have any data on which jockeys, trainers, and owners pick up on biases quickly and adjust and vice versa.

To get that refined, you'd have to watch a specific circuit closely enough to have good bias notes and be able to observe who seems to be riding aggressively to avoid a negative bias or to take advantage of a favorable one. I don't keep information like that either, but from time time I make mental notes. That's the kind of thing that's probably not useful often enough for them to care, but if you have 10 things like that you might have the beginnings of a portfolio of info with value.

Classic,

What you've mentioned is precisely what I was saying in my Global vs Local comment.

What the Whales (and anyone doing big data analysis) do right is just that: they analyze large amounts of data and they do it very well.

What they don't do well is adapt to change and variance. The idea of building models and metrics that are geared towards small models fall into the area of "local optimization."

In other words, while they are logically very aware of the outside post position bias at Turfway Park, and they might even be able to quantify that the bias gets stronger in the winter, it is probably outside the scope of their analysis to notice a shift over the last 10 days, or to notice that early horses should be penalized more than late horses when they run inside.

I (and many of my HSH users) are doing well based upon "subset data." That is, modeling subsets of local databases.

HUH?

We build specialized and specific models based not only upon track-surface-distance but also early pace pressure and individual running styles.

HUH AGAIN?

Our model looks like:

*** Track-Surface-Distance like this race
*** Pace Pressure like this race
*** What does it take for an E horse to win?
*** What does it take for an EP horse to win?
*** What does it take for an P horse to win?
*** What does it take for an S horse to win?

The whales simply CANNOT do that because it is a constant adaptation process.


Last analogy:

We're fighting an army that is much larger than us. Guerrilla warfare is the only way we can win. We have to be small (in terms of data) and fast (in terms of ability to change).


Dave

PS: re: Mule races... nothing would surprise me but I doubt it because (to my knowledge) there is no big data solution.

[classhandicapper]

We build specialized and specific models based not only upon track-surface-distance but also early pace pressure and individual running styles.

HUH AGAIN?

Our model looks like:

*** Track-Surface-Distance like this race
*** Pace Pressure like this race
*** What does it take for an E horse to win?
*** What does it take for an EP horse to win?
*** What does it take for an P horse to win?
*** What does it take for an S horse to win?
The whales simply CANNOT do that because it is a constant adaptation process.

I love it.

I'm making a huge investment of time to really tear this game apart and build personal metrics, but it's a very slow process. Even when I get something up and running well, I immediately come up with ways to make it better. In the end though, I still see the "exceptions" and know there are limits on what I can "program". That why I still feel that no matter how much the whales dominate the pools in general, they are going to keep leaving holes to exploit.