Bill Benter /Models and Overlays

[sjk]

The point I was making is that when Benter used a combination line the effect was nothing more or less than adding a cushion to the overlay requirement to cut out the borderline plays.

The part about the whales that amazes me is how they generate the volume. If you are somewhat choosy you can make money betting overlays but these days there seem to be quite a lot of races that the public bets well and there are no plays that provide the needed cushion.

[classhandicapper]

Originally Posted by Dave Schwartz

That approach will not work.

If it did, one could just use a number like BPP that produces a win rate around 30% in the top rank. Using your theory, a horse at 2/1 is break even and 4/1 should be a killer.

The problem is that the 30% hit rate goes higher or lower based upon the odds.

Same thing with trainer stats.

Or any other factor.

I'm not talking about finding a group of horses that wins x% of the time and then betting longer odds ones based on that win%. That would be the kind of mistake you make the first week you are playing horses.

I'm talking about finding inefficiencies in a particular pool or finding specific situations that are not well understood by the public. That will give you a group of horses with some negative or positive value as a starting point for looking for actual profitable plays.

More so years ago than now, but years ago you could find profitable trainer plays that would last years.

Now most of my plays are related to specific bias, pace, and race flow situations that are sometimes misunderstood by the public. Those very specific situations are the fertile ground I search for real overlays within.

[classhandicapper]

Originally Posted by sjk

The point I was making is that when Benter used a combination line the effect was nothing more or less than adding a cushion to the overlay requirement to cut out the borderline plays.

The part about the whales that amazes me is how they generate the volume. If you are somewhat choosy you can make money betting overlays but these days there seem to be quite a lot of races that the public bets well and there are no plays that provide the needed cushion.

As usual I agree with everything you are saying.

Maybe some of the whales are at the stage where they are betting some races with the expectation of very small long term losses even with rebate just to generate the necessary volume to get the overall rebate deal they get.

For example, if I put 10k through the window once in awhile where I'm expecting to lose $100-$200 long term, but those extra 10k bets allow me to meet the handle quota that gets me an extra 1% on everything else it could be worth it.

[mountainman]

Originally Posted by classhandicapper

I usually start by having a horse in mind to key on or against because of some trip, bias, trainer pattern, pace projection etc...that I think may present value. After scanning the field I have a very rough idea of where I think the horse fits. Before I pull the trigger, the odds are going to have to be screaming to me to make the play because I know I am sometimes dealing with inaccurate or incomplete information and understanding.

I call that my "margin of safety" (stolen from Ben Graham and Warren Buffett).

But I am really doing exactly what the modelers are doing when they adjust their odds lines to incorporate the public odds. They are creating a margin of safety.

This sounds like an extensive way of explaining what all good handicappers/horseplayers do. So I'm curious about why you would liken it to computer modeling?

They seek value-you seek value. But judging from your post, that's where the similarity stops.

[AndyC]

Originally Posted by classhandicapper

I'm not talking about finding a group of horses that wins x% of the time and then betting longer odds ones based on that win%. That would be the kind of mistake you make the first week you are playing horses.

I'm talking about finding inefficiencies in a particular pool or finding specific situations that are not well understood by the public. That will give you a group of horses with some negative or positive value as a starting point for looking for actual profitable plays.

More so years ago than now, but years ago you could find profitable trainer plays that would last years.

Now most of my plays are related to specific bias, pace, and race flow situations that are sometimes misunderstood by the public. Those very specific situations are the fertile ground I search for real overlays within.

Isn't an inefficiency another name for an overlay or underlay? As such how would someone determine in inefficiency if they had not determined the probability of an event happening?

[MJC922]

Originally Posted by AndyC

Isn't an inefficiency another name for an overlay or underlay? As such how would someone determine in inefficiency if they had not determined the probability of an event happening?

Yea, overlays are overlays. I suspect what the big deal is with all of this 'incorporating' is that a certain set of factors / ranks (whatever) let's say moves a 10% public odds probability a certain amount and the same set of factors moves a 50% public odds probability another possibly far greater amount altogether. So it's likely a non-linear relationship. Nobody really 'needs' to do this to make money but I suppose if you want to play in a systematic fashion using a model on a grand-scale (and many do for the potential silly amounts of money that might be made) then it makes sense to explore along these lines.

[classhandicapper]

Originally Posted by mountainman

This sounds like an extensive way of explaining what all good handicappers/horseplayers do. So I'm curious about why you would liken it to computer modeling?

They seek value-you seek value. But judging from your post, that's where the similarity stops.

I think we all agree that if you build an odds lines model you will sometimes be missing information about the horses that is built into the public odds. So some of the overlays your model flashes aren't actually overlays..

By bringing the public odds into the model, you are bringing in the information you are missing. That makes your odds line more accurate.

However, it's my opinion that the main reason guys like Benter were winning was not that they incorporated the public odds into their model.

It was the value oriented insights that were also part of their model that were the critical component.

I agree with SJK.

Adding in the public odds helped them gain a greater margin of safety in their actual bets by getting rid of some of the false overlays.

My point has been if you believe all the above is true, if you focus your attention exclusively on races where you have a value oriented insight you really don't need to build an odds line model at all. Why waste time on the other races?

[classhandicapper]

Originally Posted by AndyC

Isn't an inefficiency another name for an overlay or underlay? As such how would someone determine in inefficiency if they had not determined the probability of an event happening?

Yes.

Let's take an very simple example.

I don't want to give out specific names, but there are some trainers that have very big reputations whose record in recent years has been terrible. The public still bets them heavily off their reputations. I can show you 5 year data and 1 year data that demonstrates you would be losing massively more than the track betting on their horses. Their ROI is terrible.

For me, that's potentially something to exploit.

If I see one of their horses entered and he's going to be a relatively short price, I'm already thinking in terms of making a play against that horse if I can find something else positive the race - like some horse coming off a biased trip that people seem to have missed.

I'm not going through some complex algorithm and creating an exact odds line.

I have a horse I think is probably going to be overbet and a horse that is probably going to be underbet in the same race.

At that point all I want to know is if my positive horse is good enough to contend. If he is, I'm going to construct a bet around him while leaving the negative trainer horse off and look at the board to see if it's singing to me. I'm not to create an odds line for every horse in the race.

[Nitro]

Originally Posted by sjk

The point I was making is that when Benter used a combination line the effect was nothing more or less than adding a cushion to the overlay requirement to cut out the borderline plays.

The part about the whales that amazes me is how they generate the volume. If you are somewhat choosy you can make money betting overlays but these days there seem to be quite a lot of races that the public bets well and there are no plays that provide the needed cushion.

I believe that using a subjective interpretation (on it’s own) about the probability of any entry winning a race is the wrong approach. That’s why Benter had to enhance his statistical model to include the live odds to achieve his eventual big time success.

I have to repeat this quote from the Benter editorial:

After all, Bill Benter has written it in 1995, more than 20 years ago. There has certainly been a myriad of much more sophisticated and advanced models written in the meantime, which could easily beat Bill Benter‘s model on the racetrack.

I may not have the financial resources (yet) that Mr. Benter had, but believe it or not I can safely say that the methodology I use (which is not a "model") for playing HK (in particular) is more consistent and profitable than any model that he was using. It certainly doesn't require the work, study, and manpower that was involved in its maintenance.

http://www.paceadvantage.com/forum/s...d.php?t=161124

[Dave Schwartz]

Originally Posted by classhandicapper

I don't want to give out specific names, but there are some trainers that have very big reputations whose record in recent years has been terrible. The public still bets them heavily off their reputations. I can show you 5 year data and 1 year data that demonstrates you would be losing massively more than the track betting on their horses. Their ROI is terrible.

For me, that's potentially something to exploit.

Oh, sing it, brother!



^^^
THIS!

This is the key to success for a non-whale! (Whales have enough resources to go after the high side approach. We do not.)

Having a play-against scenario that all but automatically gives you back the takeout.

This is precisely the strategy I employ. (Not trainer-based, but finding the bet against horses.)

[mountainman]

Originally Posted by sjk

The point I was making is that when Benter used a combination line the effect was nothing more or less than adding a cushion to the overlay requirement to cut out the borderline plays.

Tx very much for elucidating. But wouldn't such a combination line raise the overlay threshold only when the public pegged a horse at higher odds (thus assigning said runner a lesser chance) than did the handicapping model????

And on a corollary note, wouldn't the opposite scenario actually LOWER the overlay threshold??

[sjk]

The way I am thinking of combining the lines is by normalizing the probabilities and making a linear combination of the two lines.

Since the probabilities are reciprocal to the odds the combination line will have less of an overlay cushion when the public probability is greater.

[classhandicapper]

It would be interesting to see which factors Benter analyzed independently and how he incorporated the public odds.

I recently had a brief conversation with someone on Twitter that developed a Benter like model. I asked if he looked at how each factor he analyzes works at different odds ranges and he said "yes".

So for all we know, Benter may not have been using public odds the same way for every horse. He may have used them as a filtering system where the standards (or margin of safety) for a play varied in different odds ranges or for different factors.

For example, maybe the margin of safety is much larger for longshots and big favorites than mid priced horses (or whatever)???

If the model spit out fair odds of 19-1, 25-1 wasn't enough. He needed 30-1.

If the model spit out fair odds of 4-1, 5-1 was enough.

If the model spit out fair odds of 3/5, 4/5 wasn't enough, 1/1 was marginal and 6/5 was a play.

[dnlgfnk]

Originally Posted by classhandicapper

Yes.

Let's take an very simple example.

I don't want to give out specific names, but there are some trainers that have very big reputations whose record in recent years has been terrible. The public still bets them heavily off their reputations. I can show you 5 year data and 1 year data that demonstrates you would be losing massively more than the track betting on their horses. Their ROI is terrible.

For me, that's potentially something to exploit.

If I see one of their horses entered and he's going to be a relatively short price, I'm already thinking in terms of making a play against that horse if I can find something else positive the race - like some horse coming off a biased trip that people seem to have missed.

I'm not going through some complex algorithm and creating an exact odds line.

I have a horse I think is probably going to be overbet and a horse that is probably going to be underbet in the same race.

At that point all I want to know is if my positive horse is good enough to contend. If he is, I'm going to construct a bet around him while leaving the negative trainer horse off and look at the board to see if it's singing to me. I'm not to create an odds line for every horse in the race.

Yes.

I don't utilize trainers much, except to fit them into the scenario whereby I am downgrading low bet horses. If First Timer is 35% to win, that leaves me with potentially 65% of negative variable information to apply to him (potentially wide turn, low trainer pct. w/ first starters, etc.).

Seems to me it can all be made more concise in the thread. Justifiably eliminate one low bet horse and one has roughly eliminated the takeout. Reasonably eliminate another horse being significantly bet and one can get a positive ROI over time.

My criteria focus upon the intra dynamics of past pace (rarely a fractional time-more so the running styles, finish positions and final figure[ high fig/speculatively good trip...low fig/bad trip]), and position on the track, when contrasted with today's potential pace and position on track.

[dnlgfnk]

Originally Posted by classhandicapper

It would be interesting to see which factors Benter analyzed independently and how he incorporated the public odds.

I recently had a brief conversation with someone on Twitter that developed a Benter like model. I asked if he looked at how each factor he analyzes works at different odds ranges and he said "yes".

So for all we know, Benter may not have been using public odds the same way for every horse. He may have used them as a filtering system where the standards (or margin of safety) for a play varied in different odds ranges or for different factors.

For example, maybe the margin of safety is much larger for longshots and big favorites than mid priced horses (or whatever)???

If the model spit out fair odds of 19-1, 25-1 wasn't enough. He needed 30-1.

If the model spit out fair odds of 4-1, 5-1 was enough.

If the model spit out fair odds of 3/5, 4/5 wasn't enough, 1/1 was marginal and 6/5 was a play.

Probably not. Reads like he played every race but one teeming with firsters.