Optimal Objective Function for Profit?

[Wizard of Odds]

Clearly maximizing winners isn't it.
Mazimizing return gives way too much weight to longshots.

So, what is the optimal objective function that won't fool you into thinking you have a winning system when you don't?

Just thinking.

[rrbauer]

So, what is the optimal objective function that won't fool you into thinking you have a winning system when you don't?

The size of your bankroll.

[twindouble]

Clearly maximizing winners isn't it.
Mazimizing return gives way too much weight to longshots.

So, what is the optimal objective function that won't fool you into thinking you have a winning system when you don't?

Just thinking.

In my opinion in todays market of pools your not just looking for winners, it's like investing in a portfoilo of pools with a wagering strategy that covers possible upsets and tossing out those horses that don't figure in the races. In other words you bet the races you like in the pools that will produce a profit or a dam good score. There's a prevailing wagering condition in every race, either there's value or there isn't, don't matter if it's one race or six races you settle on or if you have to use a chalk along the way in the picks or gimmicks.

When you get right down to it, the Optimal objective is, having the ability to evaluate your potential to make money with your handicapping skills. That incorperates a flexable wagering strategy.


Good luck,

T.D.

[GameTheory]

Clearly maximizing winners isn't it.
Mazimizing return gives way too much weight to longshots.

So, what is the optimal objective function that won't fool you into thinking you have a winning system when you don't?
Well, maximizing return IS what you want, of course -- the problem is just if you optimize for that without any "brake" against overfitting you will as you say end up with something emphasizing not only longshots, but longshots that may never recur in your lifetime. (I think I know where you are going with this, but you might want to introduce the topic a bit more so people know what you are talking about -- "objective function" is a nonsense phrase for most I would think.)

So overfitting is the problem, not the objective function per se. I don't think there is some particular objective function that you can minimize/maximize and have it automatically balance everything for you and work beautifully into the future -- if the learning algorithm is prone to overfitting, you have to deal with that no matter the objective function...

[midnight]

So, what is the optimal objective function that won't fool you into thinking you have a winning system when you don't?

Just thinking.

The optimal objective is to have an overlay when you make a bet. That may sound simplistic, but that's what it boils down to.

[akgandhi]

can you please elaborate on why you say maximizing returns gives way too much weight too longshots - I was under the impression that the well known empirical regularity of the "favorite longshot" bias shows that return from betting favorites exceeds the return from betting longshots.

[Hosshead]

So does Any method that shows a profit, consist of bets that are (on Avg.), overlays? For example, if you bet 10 even money favs, and 6 win,- then you have, on avg., bet overlays ? So the words profit and overlay are (by definition) synonymous ? (The key word here is On Avg.)

[GameTheory]

So does Any method that shows a profit, consist of bets that are (on Avg.), overlays? For example, if you bet 10 even money favs, and 6 win,- then you have, on avg., bet overlays ? So the words profit and overlay are (by definition) synonymous ?On average over a large sample, yes. You can be in the black in the short-run betting on underlays by being lucky, but in the long run if you are ahead then you must be betting (on average) overlays. (This assumes your payoffs are spread over the sample -- if you hit one monster longshot that puts you in the black then maybe we're just talking about luck again. Longshot betting would need a much bigger sample to validate statistically.)

The term overlay simply means that the horse pays more than it should given its chance to win the race. It "should" pay (on average) just enough to return exactly what you put in, minus the takeout. So even if you aren't making a profit, but are losing less than the takeout, you are also betting on overlays, just not big enough overlays.

[twindouble]

The optimal objective is to have an overlay when you make a bet. That may sound simplistic, but that's what it boils down to.

I think the words overlay and underlay are over stated. I look at it this way, what the heck do I care what others think about the race, if I like a horse I don't care if he's 100-1, 40-1 ML or 6-5, ML, going off 2-1, for that matter. When I can put together a wager that makes money in any darn pool that exists that's the direction I'm going.

T.D.

[Hosshead]

GT,
I think the key (overlay) word(s) might be "On Avg."
Lets say that the above example, with 1/1 Favs. was not just 10 races, but a large sample. In other words, somebody can select (out of all 1/1 favs), 60+% winners, of those even money favs. (I'm just using this as an extreme example)

Then you would have to say that he is betting (On Avg.) - Overlays. Right ?
So, Any profit, arrived at by Any means, (with, as you say, a large enough sample) would have to be deemed, "Betting On Overlays" ?

[michiken]

Assuming that the gambling gods are working in your favor, which is better?

a. A $8.00 winner
b. A $8.00 place price
c. A $8.00 show price

If you have a field size of 8 horses, your random odds of selecting these are:

win = 1/8 = 12.5%
place = 2/8 = 25%
show = 3/8 = 37.5%

In my opinion, the RISK ASSESSMENT is lowest for the show pool......

The objective should be maximizing your roi while having a minimum risk assessment?

[GameTheory]

I think the key word(s) are "On Avg."
Lets say that the above example, with 1/1 Favs. was not just 10 races, but a large sample. In other words, somebody can select (out of all 1/1 favs), 60+% of those even money favs. (I'm just using this as an extreme example)

Then you would have to say that he is betting (On Avg.) - Overlays. Right ?
So, Any profit, arrived at by Any means, (with, as you say, a large enough sample) would have to be deemed, "Betting On Overlays" ?Yes, I would say so. You can't make a profit if you're not betting on overlays, therefore if you are making a profit you are betting on overlays. Whether you like it or not!

[GameTheory]

Assuming that the gambling gods are working in your favor, which is better?

a. A $8.00 winner
b. A $8.00 place price
c. A $8.00 show price

If you have a field size of 8 horses, your random odds of selecting these are:

win = 1/8 = 12.5%
place = 2/8 = 25%
show = 3/8 = 37.5%

In my opinion, the RISK ASSESSMENT is lowest for the show pool......

The objective should be maximizing your roi while minimumimizing the risk assessment?
In which case the Kelly criterion could be used as objective function...

[twindouble]

Assuming that the gambling gods are working in your favor, which is better?

a. A $8.00 winner
b. A $8.00 place price
c. A $8.00 show price

If you have a field size of 8 horses, your random odds of selecting these are:

win = 1/8 = 12.5%
place = 2/8 = 25%
show = 3/8 = 37.5%

In my opinion, the RISK ASSESSMENT is lowest for the show pool......

The objective should be maximizing your roi while minimumimizing the risk assessment?

What does that have to do with handicapping the race and coming up with a 20-1 shot you like to win or you can use in the picks and make a few grand? Or, keying a 6-5 with two horses you like that will produce a $50 or $60 exacta, better yet a $1500 super.

[Hosshead]

Yes, I would say so. You can't make a profit if you're not betting on overlays, therefore if you are making a profit you are betting on overlays. Whether you like it or not!

I know we have to decide before the race but:
Therefore all winners, are (after the fact), overlays ?

[rokitman]

can you please elaborate on why you say maximizing returns gives way too much weight too longshots - I was under the impression that the well known empirical regularity of the "favorite longshot" bias shows that return from betting favorites exceeds the return from betting longshots.

There was a statistical bias towards favorites that has evaporated.

As for "maximizing returns gives way too much weight to longshots." That statement is just way too broad and open-ended.

[GameTheory]

I know we have to decide before the race but:
Therefore all winners, are (after the fact), overlays ?Only if you can pick them beforehand. (What is or is not an overlay is subjective and depends on your selection method. So even though we use the shorthand "that horse is an overlay" there is always an unstated "given", and that given is the method you use to select horses.)

Where are we going with this?

[akgandhi]

There was a statistical bias towards favorites that has evaporated.



The statistical bias may have diminished, but I find it hard to believe it has evaporated unless all bettors in the market have suddenly become risk neutral investors (quite unlikely). Is there any evidence on the change in the bias as of late?

[GameTheory]

The statistical bias may have diminished, but I find it hard to believe it has evaporated unless all bettors in the market have suddenly become risk neutral investors (quite unlikely). Is there any evidence on the change in the bias as of late?It isn't completely gone. It exists, but is weaker than it used to be.

[rokitman]

The statistical bias may have diminished, but I find it hard to believe it has evaporated unless all bettors in the market have suddenly become risk neutral investors (quite unlikely). Is there any evidence on the change in the bias as of late?

I think if you try the forum search on the matter you will find some statistics. It has been discussed here before.

It hasn't evaporated completely, as GT said. There is still a wet spot where the puddle once was ;)

[Wizard of Odds]

The problem is overfitting the historical data.

An interesting suggestion tat has some intuitive appeal is to train/evaluate the data where the amount bet equals 1/(1+odds). This does not unduely reward longshots.

Natural log (ln) of (1+odds) is another possibility.

I need to review the statistical underpinnings of the optimal strategy. Someone suggested to me the 'logit' function, but I need to look this up.

[Wizard of Odds]

Just to be clear from my last post. It's Game Theory who has most addressed the question I'm asking. <I'm new to forum and just getting used to its format>

Thanks

Well, maximizing return IS what you want, of course -- the problem is just if you optimize for that without any "brake" against overfitting you will as you say end up with something emphasizing not only longshots, but longshots that may never recur in your lifetime. (I think I know where you are going with this, but you might want to introduce the topic a bit more so people know what you are talking about -- "objective function" is a nonsense phrase for most I would think.)

So overfitting is the problem, not the objective function per se. I don't think there is some particular objective function that you can minimize/maximize and have it automatically balance everything for you and work beautifully into the future -- if the learning algorithm is prone to overfitting, you have to deal with that no matter the objective function...

[Jeff P]

Objective Function defined:
http://www.nist.gov/dads/HTML/objective.html


-jp

.

[Overlay]

To me, the optimal objective function is one that assigns proper statistical weights to a workable number of independent, non-redundant major handicapping variables in a manner which will allow you to estimate with sufficient accuracy the true winning chances of every horse in a field, without having to rely on subjective opinion. You can then confine wagers to horses and exotic-wager combinations that are going off at odds that are above fair value, wherever they may fall within the field's odds spectrum. For me, those variables encompass considerations of distance, running surface, form, speed, pace, class, track bias, and connections.

[twindouble]

To me, the optimal objective function is one that assigns proper statistical weights to a workable number of independent, non-redundant major handicapping variables in a manner which will allow you to estimate with sufficient accuracy the true winning chances of every horse in a field, without having to rely on subjective opinion. You can then confine wagers to horses and exotic-wager combinations that are going off at odds that are above fair value, wherever they may fall within the field's odds spectrum. For me, those variables encompass considerations of distance, running surface, form, speed, pace, class, track bias, and connections.


Didn't I say the same thing in different words or was I overfitting? Hold on, I'll have to compute all this to prove it out. :bang:

T.D.

[kenwoodallpromos]

Long term return.

[dancingbrave]

Optimising profits is all about betting with positive expectation everytime ie: betting at odds greater than their "real" chances. Favourites or Outsiders is irrelevant. If you don't have a handicapping system that assign accurate probabilities to horses chances of winning then all strategies are immaterial IMO.

Staking is then the issue and I would advocate a modified Kelly system. Straight Kelly is too aggressive IMO. I use fractional multiples of kelly depending of grade of race with no bet being above half Kelly and no bet being below 0.2Kelly.