Benter & public odds

[dnlgfnk]

Dave,

I searched at length unsuccessfully for a statement you may have made regarding the degree to which Bill Benter incorporated public odds into his model. If I remember correctly, public odds contributed "75 percent" towards a horse's objective odds in his model?

Apologies if I am way off on the figure or on your being the source of the quote.

[lansdale]

Originally Posted by dnlgfnk

Dave,

I searched at length unsuccessfully for a statement you may have made regarding the degree to which Bill Benter incorporated public odds into his model. If I remember correctly, public odds contributed "75 percent" towards a horse's objective odds in his model?

Apologies if I am way off on the figure or on your being the source of the quote.

You should certainly wait for Dave's answer, but I'm sure he didn't say that Benter claimed that public odds contributed 75% toward the creation of his model. If you are interested in pursuing this line, you should pick up a copy of 'Efficiency of Racetrack Betting Models' which contains the well-known article in which Benter describes his methods. Unfortunately this is no longer available online. If Magister Ludi is around, I have no doubt he has a copy of this work and might be willing to share the algorithm Benter used to combine his model's estimate with that of the public.

Cheers,

lansdale

[davew]

I have the 1994 book 'Efficiency of Racetrack Betting Markets'

Benter has an article in there 'Computer Based Horse Race and Wagering Systems: A Report'


Bender recommends 'Searching For Positive Returns at the Track: A Multinomial Logit Model For Handicapping Horse Races' by Ruth Bolton and Randall Chapman' originally in Management Science, Vol 32, No 8 August 1986 1040-1059 - but also reprinted in the Efficiency Book - as what is needed for the model.

He then compared the models prediction probability to the publics odds predictions (to see how good the model is). There always was a bias towards the publics probability - whether higher or lower than the models.

He felt it would be better if he could use another logit model using both the calculated probability AND the public odds probability as independent variables...


then he lost me ....

[Cratos]

Originally Posted by davew

I have the 1994 book 'Efficiency of Racetrack Betting Markets'

Benter has an article in there 'Computer Based Horse Race and Wagering Systems: A Report'


Bender recommends 'Searching For Positive Returns at the Track: A Multinomial Logit Model For Handicapping Horse Races' by Ruth Bolton and Randall Chapman' originally in Management Science, Vol 32, No 8 August 1986 1040-1059 - but also reprinted in the Efficiency Book - as what is needed for the model.

He then compared the models prediction probability to the publics odds predictions (to see how good the model is). There always was a bias towards the publics probability - whether higher or lower than the models.

He felt it would be better if he could use another logit model using both the calculated probability AND the public odds probability as independent variables...


then he lost me ....

Gee, I wish Magister Ludi or TrifectaMike would chime on this discussion with their opinions; I sure sure they would be enlightening.

[Magister Ludi]

Mr. Benter used a type of pseudo R^2 to assess the explanatory power of a model. For the period 1988-1993, his R^2 for the public's probability estimate was .1218. At the same time, the R^2 for his fundamental model was .1245. A model combining his fundamental model and the public's probability estimate had an R^2 of .1396. To illustrate the increasing competitiveness of this market, today's R^2 of the public's probability estimate is about .18.

[Dave Schwartz]

I have said enough on this topic already.

[dnlgfnk]

Thanks, Magister and others.

I hope I didn't mislead in that I'm not on any high-tech quest for missing information. The well known "Contingencies" article featuring Benter (May '09) revealed that he overestimated his advantage until including public odds as a factor. I was curious about the machinations which would produce "true odds" of 15 percent, when he had the subject horse at 20 percent and the public at 10 percent.

To elaborate, I spent many years over-emphasizing the value of difficult trips, largely due to Beyer's relating the successes of the visual handicapper "Charlie" in a few of his works. Where the public might be impressed by a big figure that denoted ability, I speculated long ago that a bad figure meant a bad trip (except for obvious, never-pick-up-their-feet types who go off around 40-1 or more), and a big figure meant an easy trip. In many cases it still does for me, particularly with favorites who will only win 35% of the time or less. I would scrutinize the charts, plotting positions at each call before the stretch, looking for wide horses on the turn, rail speed that habitually breaks behind leaving the chute and rushes up, and premature energy spent (dueling with the favorite, too far back of lone big speed and apparently urged to reach contention early on backstretch, etc.) Anything plausible to explain those poor figures and upgrade the delicious odds horses.

The past few years, I have become much more attuned to the percentages in the game as established by the public (thus the incorporation of public odds as a factor, i.e., Benter), and it's not so much the impossible recent trips that make a horse 30-1, as it is the solid group between the favorite(s) and deserved longshots. And since those public percentages are achieved significantly by the use of conventional, obvious information, I have returned to more convention, or as Ernie Dahlman would say, becoming a Master of the Obvious.

[Sapio]

Originally Posted by Magister Ludi

Mr. Benter used a type of pseudo R^2 to assess the explanatory power of a model. For the period 1988-1993, his R^2 for the public's probability estimate was .1218. At the same time, the R^2 for his fundamental model was .1245. A model combining his fundamental model and the public's probability estimate had an R^2 of .1396. To illustrate the increasing competitiveness of this market, today's R^2 of the public's probability estimate is about .18.

Magister, does that mean that most models are ineffective in view of the publics probability estimate of .18?

Thomas Sapio

[Magister Ludi]

Originally Posted by Sapio

Magister, does that mean that most models are ineffective in view of the publics probability estimate of .18?

Thomas Sapio

Yes. This financial market is getting more and more efficient. However, there are inefficiencies that can be successfully exploited.

[Cratos]

Originally Posted by Magister Ludi

Mr. Benter used a type of pseudo R^2 to assess the explanatory power of a model. For the period 1988-1993, his R^2 for the public's probability estimate was .1218. At the same time, the R^2 for his fundamental model was .1245. A model combining his fundamental model and the public's probability estimate had an R^2 of .1396. To illustrate the increasing competitiveness of this market, today's R^2 of the public's probability estimate is about .18.

Thanks for the explanation.

[classhandicapper]

I don't build models like this, but IMO the significance of public betting is dependent on the race.

If it's a race with several first time starters, significant layoffs, suspicious drop downs etc.. the public money could reflect information about the horse's condition that an outsider wouldn't have access to.

If it's the Kentucky Derby or Breeder's Cup, there's an endless supply of unsophisticated money. So I'd totally ignore anything that doesn't make sense.

I'm sure incorporating the public odds improves these kinds of models, but I've always considered having to do that as a weakness in the model.

I consider it a weakness in my own handicapping!

When I make a line that strongly disagrees with the public line and I can't explain it, if the public horse wins, I always go back to the race to see if I missed something or can learn something.

The public is so darn good, that unless I can explain why I'm getting good value (I have unique information or a unique understanding of the situation), I usually just assume it's me that's missing something or misunderstanding something. But that's different than incorporating their estimate of the odds into my thinking all the time. If I know exactly why the odds are the way they are and disagree, that's when I want to pull the trigger.

[Magister Ludi]

Roughly 10% of the thoroughbred pari-mutuel handle is invested in horses that are perceived to be overlays. The remaining 90% of the handle is bet on horses that are perceived to be “contenders”, or potential winners. It is the 10% minority that makes the market efficient - not the 90% majority.

[classhandicapper]

Originally Posted by Magister Ludi

Roughly 10% of the thoroughbred pari-mutuel handle is invested in horses that are perceived to be overlays. The remaining 90% of the handle is bet on horses that are perceived to be “contenders”, or potential winners. It is the 10% minority that makes the market efficient - not the 90% majority.

That makes a lot of sense to me. That's why it's so tough. The smartest 10% is correcting the errors of the other 90%.

But I consider "insiders" with knowledge that is not available to me in specific instances to be part of the 10% and also think on big stakes days it might be 3% to 97% (or whatever) and that changes everything.

I guess what I am saying is that the value of the public line is a variable and knowing that can help you get into the 10% given that some models may presume it's a constant.

[delayjf]

To elaborate, I spent many years over-emphasizing the value of difficult trips, largely due to Beyer's relating the successes of the visual handicapper "Charlie" in a few of his works.

I can’t help but wonder; just as Beyer has changed the way he applied his speed figures back from when that book was written, so to has the application of trip handicapping changed.

Back in the 70’s with the public masses not using speed figures, Beyer could make a profit focusing on those horses with the best last race speed figures. In that same light, I’m guessing that Charlie was able to focus on the kinds of bad trips that were ignored then, but are now more obvious and available to the public.

It would be interesting to hear from those who used trip handicapping back in its infancy to see if what worked then, works today or what kind of trips they look for today compared to the good old days. Maybe THELITTLEGUY could comment, as I believe he’s big into trip handicapping.

I’ve a sneaking suspicion that some trainers, in an attempt to stay ahead of the crowd, have changed their SOP when running back horses that have had obvious and well documented bad trips. After observing and betting several “bad trip” horse some years back, I started to see a pattern whereby the bad trip horse did not run an improved race in its next start after the bad trip, Rather they showed the anticipated improvement in the second start after the obvious bad trip. Leaving me to believe that the first race back was used as a kind of conditioner to maintain or improve their fitness, while at the same time throwing the informed public off the scent. Then go for the money in the second race back at higher odds.

[Sapio]

Originally Posted by Magister Ludi

Roughly 10% of the thoroughbred pari-mutuel handle is invested in horses that are perceived to be overlays. The remaining 90% of the handle is bet on horses that are perceived to be “contenders”, or potential winners. It is the 10% minority that makes the market efficient - not the 90% majority.

Am I safe to assume that the majority, 90%, are betting on winners and not value? If so, not much has changed over the years. I believe I read on this forum that the shape of the odds distribution has remained the same over the years. I didn't grasp what that meant at the time. Thanks for the clarification.

Thomas Sapio