Random

[karlskorner]

NetCapper's Weekly Racing Quote from Steve Fierro's book "The Four Quarters of Horse Racing"

"We tend to lose sight of the fact that 20% of what goes on in racing can't be explained. In fact, if people really knew how random the game is they would quit (well, probably not)"

I am more inclined to believe that "50%" of what goes on in racing can't be explained.

Karl

[Que]

I would love to be able to explain 80% of what goes on in horse racing... but, I think I have a long way to go to make that happen. In the meantime, I'll settle for understanding just 20%-- especially if the odds are right.

Que.

[karlskorner]

My thoughts on Random. When I turn to the next race, it never happened before and it will never happen again. Now thats Random.

Karl

[anotherdave]

Years ago when I tried multiple linear regression techniques, it was hard for me to get much more than 20 or 25% "explained" (r-squared for the statistical minded). And at that I was breaking even or making a small profit. If we could explain 50%, we'd all be rich.

AD

[Rick]

AD,

You can do better than that if you include actual odds but I think the highest r squared I ever got was about .35 for predicting finish position. It may be possible to get up to around 40% by adding a large number of factors though (I only used 3 of my own). The Hong Kong guys probably can do that. The only information I have on them was for using logit regression to predict wins, and the numbers for that approach are not comparable with those for predicting finish position. In case anyone was wondering, morning line odds did not add anything to the prediction in my studies.

[anotherdave]

Good point, Rick. I imagine that using final odds would get you up in the 40% bracket. I never used them in my studies as at the time. I would have, but I was unable to get to the track - just phone bets and the internet wasn't available to give me all the information. So I was trying to do it with the info in the form.

AD

[cato]

I first heard the concept of 20% random from Mitchell (what's happened to him anyway?). I think his explanation was that you should be able to get the winner in your top 4 or 5 contenders 80% of the time; and in that sense the game is predictable. The other 20% of the time the winner closes from Mars.

So in putting together a line, Mitchell would assign 80% to his contenders and 20% to the rest of the field. I think that Cramer does the same thing.

Within that context, I think the description is fairly accurate.

But to say that 80% of racing is logical or that racing is 80% logical or predictable is, well, illogical....

Regards, Cato

[rrbauer]

Rick wrote:
In case anyone was wondering, morning line odds did not add anything to the prediction in my studies.

Comment/question:
Were your studies' circuit/track/race-type/field-size specific? The reason that I ask is that I have done studies of the SoCal circuit that involved ML odds and have found that there is a positive correlation between ML odds' rank and finish position. ML rank is simply ML fav, ML 2nd fav, ML 3rd fav, etc. As an aside, at the time (and I used three years' data) there were two ML makers on the circuit's major tracks (Tufts doing SA/DMR and Wyatt doing HOL). The differences in correlation were not significant.

[Rick]

rrbauer,

It's true that morning line odds itself is a significant predictor of finish position. But if you combine it with actual odds you find that it doesn't add anything to just using actual odds alone. In other words, it's already fully incorporated in actual odds. In fact, about 80% of the variance in actual odds is explained by morning line odds alone. The studies I did were on Southern California tracks a couple of years ago, and since the morning line is pretty strong there, I expected to find some significance. You might very well find it if you looked at just predicting winners alone though, because I have seen periods of time when the morning line favorite there did much better than expected.

[rrbauer]

The reasons that I used ML odds:

1. It's one thing to use data from a chart, after the fact, and have it be a constant factor. It's a different story when you have to deal with the dynamics of the realtime betting environment. It's impossible to know final odds when you bet and as we've chronicled here in the past, that is a logistical problem that has gotten worse with expanded simulcasting.
2. When you're playing multi-race "pick" bets, only the leadoff leg gives visibility of "current" odds. So, as a factor for predictability, ML is useful if it is found to be a valid predictor.

[Rick]

rrbauer,

Yeah, I like using morning line odds as a predictor of actual odds for the reasons you mentioned. If I were betting using the toteboard though, I think I'd use earlier odds, like maybe 10 minutes to post because late money tends to be smarter than early money and I'd prefer to calculate my overlays based on dumb money. But it's hard to generalize because there are some good toteboard methods based on the opposite logic. It's hard to understand the contradictions in horse racing sometimes.