Variance in exotic betting

[lansdale]

One of the discoveries of William Benter, which has been confirmed by others, was a mistake in the Harville formula in using win probability as correlated with place and show probability. As he found, the contests for place and show demonstrate far more variance than was previously thought. This is consistent with my experience - I've found that the place probability for various factors I've tested for is nearly exactly half of its win probability. I know that many people bet equal amounts on horses to win and show, although Dick Mitchell and I believe Jeff P. have explained why this is low if not zero EV in the long run. Benter's comments seems to suggest why place betting, at least of your likely winners, may be a bad idea, and why seemingly bad horses might be better plays in exotics. Any feedback appreciated.

Cheers,

lansdale

[Overlay]

I've come to appreciate Mitchell's thoughts on place-horse types, rather than viewing place probability as a strict function of individual win probabilities. However, I do also think that the more comprehensive and properly balanced/weighted a win-probability model is, the greater validity it will have as a tool for basing place-probability projections, since there will be more obstacles that low-ranking horses would have to overcome to finish at or near the front of the field.

[CBedo]

Makes sense to me. It's logical that the best win horse is not necessarily the best 2nd place horse, but since "to place" means first or second, obviously the probability of winning plays a large role in the place probability, especially if the probability of coming in second has more horses more tightly grouped with more varaince.

[fmolf]

Makes sense to me. It's logical that the best win horse is not necessarily the best 2nd place horse, but since "to place" means first or second, obviously the probability of winning plays a large role in the place probability, especially if the probability of coming in second has more horses more tightly grouped with more varaince.
this is why mitchell recommends place bets on low odds favorites as good bets.My favorite place type horse is the presser or closer to get up after the speed horses have lost the battle.the strongest survives and the pressers clean up the other two spots

[Overlay]

this is why mitchell recommends place bets on low odds favorites as good bets.

Tom Ainslie also once touted such bets as producing an overall 5% profit (although I don't know if that's still true or not) when discussing ways for the new player to begin improving on the rate of loss that Ainslie said would result from random betting (equivalent to the combined percentages of take and breakage).

[TrifectaMike]

One of the discoveries of William Benter, which has been confirmed by others, was a mistake in the Harville formula in using win probability as correlated with place and show probability. As he found, the contests for place and show demonstrate far more variance than was previously thought. This is consistent with my experience - I've found that the place probability for various factors I've tested for is nearly exactly half of its win probability. I know that many people bet equal amounts on horses to win and show, although Dick Mitchell and I believe Jeff P. have explained why this is low if not zero EV in the long run. Benter's comments seems to suggest why place betting, at least of your likely winners, may be a bad idea, and why seemingly bad horses might be better plays in exotics. Any feedback appreciated.

It is not that the Harville formula has a "mistake". Since Harville uses win probabilities it is adversely affected by a favorite-longshot bias: favorites tend to be underbet while longshots tend to be over bet. Some have made attempts at correcting the bias, but I'm not convinced they have been
successful.

A far more superior method is the exacta pools to estimate exacta probabilities. The implication is that if you're trying to gain from the exacta pool by using win probabilities, it's futile. But there is a silver lining. You can use the exacta pool probabilities to take advantage of the win probabilities.

Mike

[Overlay]

A far more superior method is the exacta pools to estimate exacta probabilities. The implication is that if you're trying to gain from the exacta pool by using win probabilities, it's futile. But there is a silver lining. You can use the exacta pool probabilities to take advantage of the win probabilities.

As you say, exacta probabilities (according to the betting public) can be calculated from the exacta pools. But isn't the point to find out where those probabilities are incorrect or inefficient compared to the "actual" probability (however you choose to calculate it) of each combination coming in, the same as it would be with trying to find overlays in the win pool by comparing the toteboard odds for each horse with your estimation of the horse's true winning chances?

[Overlay]

Tom Ainslie also once touted such bets as producing an overall 5% profit (although I don't know if that's still true or not) when discussing ways for the new player to begin improving on the rate of loss that Ainslie said would result from random betting (equivalent to the combined percentages of take and breakage).

I went back and checked the actual quote (which was in Ainslie's Complete Guide to Thoroughbred Racing), and, according to Ainslie, the 5% profit figure applied (at that time) to place bets on odds-on favorites in major handicaps, stakes races, and featured allowance races at major tracks. (I just wanted to clarify the restrictions that Ainslie applied, for the sake of accuracy.)

[TrifectaMike]

As you say, exacta probabilities (according to the betting public) can be calculated from the exacta pools. But isn't the point to find out where those probabilities are incorrect or inefficient compared to the "actual" probability (however you choose to calculate it) of each combination coming in, the same as it would be with trying to find overlays in the win pool by comparing the toteboard odds for each horse with your estimation of the horse's true winning chances?

Perhaps what I said wasn't clear enough. Using the exacta pools is a better predictor than Harville. So, if you use Harville to attempt to gain an edge, it's futile.

[lansdale]

I've come to appreciate Mitchell's thoughts on place-horse types, rather than viewing place probability as a strict function of individual win probabilities. However, I do also think that the more comprehensive and properly balanced/weighted a win-probability model is, the greater validity it will have as a tool for basing place-probability projections, since there will be more obstacles that low-ranking horses would have to overcome to finish at or near the front of the field.

Overlay,

I somewhat regret throwing Mitchell's name out there because I think it obscured the basic question, but I think you were able to blow away most of the smoke. Benter's specific point is that the Harville formula overestimates place probablility as a function of win probability - I think others in this thread are aware of it. I also described the place probability as being eerily similar in proportion to my win probability for all tested contenders - 150% of win probability - if win is .3, place is .45. Clearly, as you say, there is a relationship, and I would be interested in knowing if others have gotten similar results or widely different. But, to me, the more interesting implication of Benter's comments is how horses with little apparent value according to our weighting of factors, still manage to hit the board. I have head-scratching experiences like this every day, and I assume others may too. Or, if not, are they using a different set of factors for win and place?

Cheers,

lansdale

[lansdale]

It is not that the Harville formula has a "mistake". Since Harville uses win probabilities it is adversely affected by a favorite-longshot bias: favorites tend to be underbet while longshots tend to be over bet. Some have made attempts at correcting the bias, but I'm not convinced they have been
successful.

A far more superior method is the exacta pools to estimate exacta probabilities. The implication is that if you're trying to gain from the exacta pool by using win probabilities, it's futile. But there is a silver lining. You can use the exacta pool probabilities to take advantage of the win probabilities.

Mike

TM,

I don't think anyone disagrees that Harville is off, however, as Overlay points out, we're still back to the problem of finding exacta pool inefficiencies, and you don't really answer that question. But, as I said above, the question I'm really asking, is, given that win probability is a relatively weak predictor of place probability, and that, as Benter says, the contests for place and show are relatively random, is the difference between a horse that wins and places, more one of degree (the place horse simply has less of the value-weighted factors) or one of difference. Or to put it another way, is their exacta wager value inherent in not having the factors that are valued in 'win' horses?

[lansdale]

Makes sense to me. It's logical that the best win horse is not necessarily the best 2nd place horse, but since "to place" means first or second, obviously the probability of winning plays a large role in the place probability, especially if the probability of coming in second has more horses more tightly grouped with more varaince.

Chris,

I'll use an example to try to illustrate the question I've been asking in this thread - if my top four contenders place only 45% as a group (that is my current rate), which is exactly half their win rate, can the other 55% of place horses be assigned equal 'random' odds, or do they have specific characterstics that can be weighed as factors for 'place' but not 'win'. Hope that's clear. BTW, thanks for your many insightful posts here.

Cheers,

lansdale

[Overlay]

Is the difference between a horse that wins and places, more one of degree (the place horse simply has less of the value-weighted factors) or one of difference. Or to put it another way, is their exacta wager value inherent in not having the factors that are valued in 'win' horses?

I think Mitchell accounts for both of those possibilities by his inclusion of unbalanced horses among his place-horse types -- the kind of horses that may be eliminated by a typical "pick-the-winner" orientation because of a flaw in one of the basic aspects of their records, or a strong bias that they are fighting, but which have enough strength in one or more other qualities to almost get the job done (which to me would be "difference"); and also by including the favorite, which finishes second more often than any other single horse if it doesn't win (which would fall under the heading of "degree"). I think that the "difference" horses would be the ones more likely to produce value in an exacta combination.

[lansdale]

I think Mitchell accounts for both of those possibilities by his inclusion of unbalanced horses among his place-horse types -- the kind of horses that may be eliminated by a typical "pick-the-winner" orientation because of a flaw in one of the basic aspects of their records, or a strong bias that they are fighting, but which have enough strength in one or more other qualities to almost get the job done (which to me would be "difference"); and also by including the favorite, which finishes second more often than any other single horse if it doesn't win (which would fall under the heading of "degree"). I think that the "difference" horses would be the ones more likely to produce value in an exacta combination.

Overlay,

I think we agree on the generalities, but what I'm trying to get at is how to assign place odds to the 'difference' or 'non-win probability' types in figuring exacta pools, since I already know that my 'degree' or 'win probability' types can be safely assigned 45% of place probability. Do I randomly split the remaining 55% of place probability among the 'difference' horses, or have you or anyone else found a way to rank the 'difference' characteristic?

Cheers,

lansdale