Bill Benter /Models and Overlays

[formula_2002]

Originally Posted by formula_2002

Emperor Joseph 2nd may have told Mozart "there are simply too may notes"

I work with numbers and not enough of them, but here are some numbers for the top three used in the exacta. they are the top three public choices at about 1 minuet to post.
in a 27 race example one of the top three won 67% of the time.
a flat win bet on all three returned a loss of $0.46 on the dollar

when the top three are used over my selected horses the positive return was $0.07.
of course the results for the next 27 races will be more interesting.

here are interesting results for the "next 27 races"

in this 27 race example one of the top three won 81% of the time.
a flat win bet on all three returned a profit of $0.04 on every 3 dollars

when the top three are used over my selected horses the positive return was $0.59 0n every three dollars

[classhandicapper]

Does anyone know if Benter used the morning line to try to separate public money from inside money?

The more I think about this the less it makes sense unless he had different margins of safety for horses in different odds ranges like I suggested previously or he was doing something to try to separate inside money from public money.

[Jeff P]

Someone once told me he asked Benter that very same question.

He said the answer he got back was along the lines of We tested using the morning line and quickly abandoned the idea because the tote worked much better.

Fyi, the above is third hand info. (I've never asked Benter that question.)

That said --

I'm someone who has tested using both the mline and the tote in a model.

While it's true the mline can be useful at times depending on the factors in your model:

My own testing suggests the tote provides significantly more info than does the mline.



-jp

.

[classhandicapper]

Originally Posted by Jeff P

Someone once told me he asked Benter that very same question.

He said the answer he got back was along the lines of We tested using the morning line and quickly abandoned the idea because the tote worked much better.

Fyi, the above is third hand info. (I've never asked Benter that question.)

That said --

I'm someone who has tested using both the mline and the tote in a model.

While it's true the mline can be useful at times depending on the factors in your model:

My own testing suggests the tote provides significantly more info than does the mline.

-jp

.

I agree with everything you are saying in terms of making a more accurate odds line, but I'm not sure how the live odds are helping him find overlays.

Let's say my own odds line gives a horse 50% and the tote odds give him 25%. Just for fun, let's split the difference and call it 37.5%.

He's an overlay either way. It's just the degree that changes.

The only thing that makes sense is if you change the margin of safety depending on the odds range. Maybe if the live odds have the horse at 20-1, I need 30-1 to make a play but if he's 2-1 I can take 3-1. That would tend to weed out some of the theoretical smaller overlays and some of the long shot overlays where it might be harder to create an accurate line.

I'm just thinking out loud.

[Poindexter]

Originally Posted by classhandicapper

I agree with everything you are saying in terms of making a more accurate odds line, but I'm not sure how the live odds are helping him find overlays.

Let's say my own odds line gives a horse 50% and the tote odds give him 25%. Just for fun, let's split the difference and call it 37.5%.

He's an overlay either way. It's just the degree that changes.

The only thing that makes sense is if you change the margin of safety depending on the odds range. Maybe if the live odds have the horse at 20-1, I need 30-1 to make a play but if he's 2-1 I can take 3-1. That would tend to weed out some of the theoretical smaller overlays and some of the long shot overlays where it might be harder to create an accurate line.

I'm just thinking out loud.

My guess is that the emphasis on his underlays, ironed out his line, making it more accurate. All underlays are not the same. If I make a horse 3-1 and he is 3/5 and I used your split the difference theory I would come up with .25 and .625/1.16 or .54 that would give me a estimated probability of .395%. I find that these horses win much closer to the .54 % than the 39.5 %. I haven't done the same research with a 20-1 at 4-1 or 10-1 at 3-1 or a 50-1 at 18-1 so I don't know what the right answer is with these horses. These teams I am sure have the right answer to every one of these situations so they are making their line better in each instance.

Also you have the other direction, does a 6-1 at 40-1 win as often as a 6-1 at 20-1 or a 6-1 at 10-1?


You answer all these questions over 100's or 1000's of races and you make your line better. The ultimate goal is the most accurate line possible, right?

[thaskalos]

Benter noticed that the tote-board favorites were beating his favorites. Being of sound mind, he put his own ego aside, incorporated the tote-board into his calculations...and became a mega-millionaire.

I too noticed that the tote-board favorites were beating my favorites. But my ego convinced me that I would eventually come up with a better odds-line than that of the seemingly unsophisticated betting public. So, I kept at it...and remained broke for the next 20 years.

[FakeNameChanged]

Originally Posted by thaskalos

Benter noticed that the tote-board favorites were beating his favorites. Being of sound mind, he put his own ego aside, incorporated the tote-board into his calculations...and became a mega-millionaire.

I too noticed that the tote-board favorites were beating my favorites. But my ego convinced me that I would eventually come up with a better odds-line than that of the seemingly unsophisticated betting public. So, I kept at it...and remained broke for the next 20 years.

Thanks for the laugh, and for bringing some realism to this long thread.

[classhandicapper]

Originally Posted by Poindexter

My guess is that the emphasis on his underlays, ironed out his line, making it more accurate. All underlays are not the same. If I make a horse 3-1 and he is 3/5 and I used your split the difference theory I would come up with .25 and .625/1.16 or .54 that would give me a estimated probability of .395%. I find that these horses win much closer to the .54 % than the 39.5 %. I haven't done the same research with a 20-1 at 4-1 or 10-1 at 3-1 or a 50-1 at 18-1 so I don't know what the right answer is with these horses. These teams I am sure have the right answer to every one of these situations so they are making their line better in each instance.

Also you have the other direction, does a 6-1 at 40-1 win as often as a 6-1 at 20-1 or a 6-1 at 10-1?


You answer all these questions over 100's or 1000's of races and you make your line better. The ultimate goal is the most accurate line possible, right?

This is more along the lines of what I am thinking.

He must have been using a different adjustment for different odds ranges. Just incorporating the public odds the same way all the time will help you make a better odds line, but it's not going to change much in terms of who you will bet.

[classhandicapper]

Originally Posted by thaskalos

Benter noticed that the tote-board favorites were beating his favorites. Being of sound mind, he put his own ego aside, incorporated the tote-board into his calculations...and became a mega-millionaire.

I too noticed that the tote-board favorites were beating my favorites. But my ego convinced me that I would eventually come up with a better odds-line than that of the seemingly unsophisticated betting public. So, I kept at it...and remained broke for the next 20 years.

I tried a different approach after I noticed the following.

If I didn't understand why the public was making a horse the favorite it won more often than the horse I thought should be the favorite.

If I did understood why the public was making the horse the favorite but had good reason to disagree, I was right more often.

I didn't makes millions or go broke.

[Jeff P]

Originally Posted by classhandicapper

I agree with everything you are saying in terms of making a more accurate odds line, but I'm not sure how the live odds are helping him find overlays.

Re: the bolded part of the above quote --

Imo, a well constructed fundamental model is the combination of two parts:

Part I. A before the odds are known model.

In his published work Benter specifically emphasized adjusted speed figs and adjusted finish positions.

Years later, during an interview, Dana Parham said their fundamental model was also scoring rider, trainer, and post position.

I've been told privately by players not affiliated with Benter in any way shape or form that their own before the odds are known models include 'trackwork' or subjective scoring of workouts and trips.

I get a very strong sense everything Benter and others like him have been willing to share barely scratches the surface. And I fully appreciate why.

I wouldn't be at all surprised if the before the odds are known part of not just Benter's model, but the models in use by a lot of the whales include many areas of the game: ability from speed figs, preference for surface and distance, class, current form, workouts, trips, early-late and inside-outside path biases, rider, trainer, breeding, and horse physicality, etc.


Part II. Win likelihood from tote.

You can write a reasonably good win likelihood algorithm using a pct of money bet on each horse in the win pool approach.

This is what that looks like in my current database over the past year:

Data Window Settings:
Connected to: C:\JCapper\exe\JCapper2.mdb
999 Divisor  Odds Cap: None
SQL UDM Plays Report: Hide

SQL:  SELECT * FROM STARTERHISTORY
      WHERE [DATE] >= #10-29-2019#
      AND [DATE] <= #10-28-2020#
      ORDER BY [DATE], TRACK, RACE


Data Summary          Win         Place          Show
-----------------------------------------------------
Mutuel Totals   368497.10     368385.30     366359.30
Bet            -485900.00    -485900.00    -485900.00
-----------------------------------------------------
P/L            -117402.90    -117514.70    -119540.70

Wins                31122         62003         90867
Plays              242950        242950        242950
PCT                 .1281         .2552         .3740

ROI                0.7584        0.7582        0.7540
Avg Mut             11.84          5.94          4.03

By: SQL-F05 Gap (Implied Public Prob from Win Pool)

  >=Min        < Max        P/L        Bet        Roi    Wins   Plays     Pct   Impact
--------------------------------------------------------------------------------------
 0.0000       0.0250  -48412.70  118574.00     0.5917    1048   59287   .0177   0.1380
 0.0250       0.0500  -16799.00   81872.00     0.7948    2077   40936   .0507   0.3961
 0.0500       0.0750  -11415.00   55474.00     0.7942    2224   27737   .0802   0.6259
 0.0750       0.1000   -7233.30   40468.00     0.8213    2225   20234   .1100   0.8584
 0.1000       0.1250   -6741.60   30756.00     0.7808    1970   15378   .1281   1.0000
 0.1250       0.1500   -5127.30   24392.00     0.7898    1844   12196   .1512   1.1803
 0.1500       0.1750   -3355.70   19732.00     0.8299    1781    9866   .1805   1.4092
 0.1750       0.2000   -2873.60   16120.00     0.8217    1599    8060   .1984   1.5487
 0.2000       0.2250   -2275.20   13622.00     0.8330    1496    6811   .2196   1.7146
 0.2250       0.2500   -1878.00   11618.00     0.8384    1390    5809   .2393   1.8679
 0.2500       0.2750   -1769.50    9590.00     0.8155    1194    4795   .2490   1.9439
 0.2750       0.3000   -1433.20    8068.00     0.8224    1085    4034   .2690   2.0996
 0.3000       0.3250   -1114.80    7014.00     0.8411    1025    3507   .2923   2.2816
 0.3250       0.3500    -590.20    5980.00     0.9013     987    2990   .3301   2.5769
 0.3500       0.3750    -957.90    5204.00     0.8159     824    2602   .3167   2.4721
 0.3750       0.4000    -649.50    4404.00     0.8525     763    2202   .3465   2.7049
 0.4000       0.4250    -618.00    3996.00     0.8453     713    1998   .3569   2.7858
 0.4250       0.4500    -349.20    3440.00     0.8985     682    1720   .3965   3.0953
 0.4500    9999.0000   -3809.20   25576.00     0.8511    6195   12788   .4844   3.7817

Although it's not reflected in the above data sample, you can improve the accuracy a little by including pct of money bet on each horse from other pools such as exa, dd, p3, and p4, etc.


Imo, where Benter, Woods, and others really separated themselves from everyday horseplayers was using advanced stat tools to combine the before the odds are known part with the win likelihood from tote part into a fundamental model.

A strong fundamental model combined with CRW allows them to:

Identify and bet likely overlays instantly just before race off.

Not just in the win pool, but in every pool.

And with proper ticket structure based on the preference order or the way their fundamental model ranks out the horses in each field.

And with optimal bet sizing for individual wagers given estimated pool size.

Keep in mind, unlike the everyday horseplayer, players or teams with a good fundamental model aren't limited to spot plays or angles.

Any horse that steps into a starting gate can be an overlay if the price is right.

If a field of eight is facing up to the gate and the 5th and 7th ranked horses in their preference order are both obvious overlays based on current odds and their fundamental model:

They are betting those two horses down. Instantly and according to formula.

If in the same race their fundamental model sees 3 underbet exacta combinations, 2 underbet trifecta combinations, and 4 underbet double combinations involving the current favorite:

They are betting those specific combinations according to formula.

Imo, what they are doing is way beyond what everyday players are capabable of.

They are making an accurate EV or expected value calculation just before race off.

Everything above break even point (with rebate factored in) gets bet.

Everything not above break even point gets passed.




-jp

.

[classhandicapper]

Jeff,

Thanks for that detailed explanation.

I understand how they are using tote information to gain an advantage finding overlays within all the pools/combinations late in the betting.

I understand how blending the real odds with a model can yield a more accurate odds line than the model alone.

What I don't really understand how that more accurate odds line translates into an advantage.

Let's say my home made model calculates that a horse should be 2-1 and he's on the board at 6-1.

My model says he's a huge overlay and I should play him.

Now let's say I blend the live odds in some way based on experience and it tells me a more accurate odds line on that horse is 4-1 not 2-1.

Either way, I'm going to play the horse. So what was the benefit?

The only thing I can think of is having data that suggests your model does better finding overlays in certain odds ranges than others or where blending the model odds and live odds reduces the theoretical overlay by enough that you pass the bet because there's not enough margin of safety in the bet.

[traveler]

Originally Posted by classhandicapper

Jeff,

Thanks for that detailed explanation.

I understand how they are using tote information to gain an advantage finding overlays within all the pools/combinations late in the betting.

I understand how blending the real odds with a model can yield a more accurate odds line than the model alone.

What I don't really understand how that more accurate odds line translates into an advantage.

Let's say my home made model calculates that a horse should be 2-1 and he's on the board at 6-1.

My model says he's a huge overlay and I should play him.

Now let's say I blend the live odds in some way based on experience and it tells me a more accurate odds line on that horse is 4-1 not 2-1.

Either way, I'm going to play the horse. So what was the benefit?

The only thing I can think of is having data that suggests your model does better finding overlays in certain odds ranges than others or where blending the model odds and live odds reduces the theoretical overlay by enough that you pass the bet because there's not enough margin of safety in the bet.

Yes, blending in the tote could erase your edge totally so you should pass. Also, if you are betting using optimal bet strategy, your bet size is a function of your edge. So if line smoothing reduces your edge, you should bet less. At least that's how I understand it. YMMV

[JerryBoyle]

Originally Posted by traveler

Yes, blending in the tote could erase your edge totally so you should pass. Also, if you are betting using optimal bet strategy, your bet size is a function of your edge. So if line smoothing reduces your edge, you should bet less. At least that's how I understand it. YMMV

Exactly right. Blending w the market can either keep you from betting if you have no edge, or reduce your bet size because your edge is smaller than you think.

[Nitro]

Originally Posted by classhandicapper

Jeff,

Thanks for that detailed explanation.

I understand how they are using tote information to gain an advantage finding overlays within all the pools/combinations late in the betting.

I understand how blending the real odds with a model can yield a more accurate odds line than the model alone.

What I don't really understand how that more accurate odds line translates into an advantage.

The only thing I can think of is having data that suggests your model does better finding overlays in certain odds ranges than others or where blending the model odds and live odds reduces the theoretical overlay by enough that you pass the bet because there's not enough margin of safety in the bet.

I hate being the bearer of bad tidings but after viewing a number of posts (particularly Jeff’s) I can see now why this co-mingling of subjectively generated pre-race odds and the actual tote information has become such a difficult issue. The problem as I see it is coming from the perspective of the handicapping data used to establish some pre-race odds based on any number of weighted variables. Then of course these same pre-race odds will have some sort of favorable or negligible weighted value. How that value is proportionately blended with the real time odds or tote activities seems to be the stickler (At least from the same perspective).

Based on what I’ve reading here, I have a funny suspicion that the blending formula that many apply is biased toward the pre-race odds rather than the actual odds. I can safely say that the objectively determined live odds far out weight the subjectively produced pre-race odds. Why? Because you’re dealing with real time money that’s being bet in an effort to make a profit. In case anyone has missed it, this game a ALL about the MONEY in so many respects. How you acknowledge and value it (be it yours or someone else’s) is the key to making rational betting decisions.

[classhandicapper]

Originally Posted by traveler

Also, if you are betting using optimal bet strategy, your bet size is a function of your edge. So if line smoothing reduces your edge, you should bet less. At least that's how I understand it. YMMV

Thanks. I forgot about that dynamic because I've never been able to use it effectively in my own betting. I more or less have two type of bets. One is "serious" and the other is "I'm so bored out my mind I'm going to $2-$5 for fun".

Among the "serious" bets I don't seem to have the requisite skill to know how much of an edge I have on any single bet. I have enough to know to let the moths out of my wallet. So my bets are very flat. I wish I could be more accurate, but I can't - even after all these decades of playing and studying my results.