Dutching by removing underlaid long shots.

[teddy]

ANyone try this idea... I have once before and got my hat handed to me. Makes sense in theory. Harder in practice. Ideally you have to safely remove 20% of the pool with removing underlaid long shots or other horses that are overbet. I start out by finding races with multiple underlaid horses..

Mtn race 3 on sun.. Removing the 1 and 3, which were underlaid odds could effectively change the takeout from 18% to about 13%...making the game a bit easier.

Key is to find more overlaid horses in the race and thus hopefully an underlay.

the 2 won and was 5-1 on my line and went off at 7-1.

[therussmeister]

I played with it about 20 years ago. The problem with dutching these days is late odds movement.

[Ocala Mike]

That would never work for me. In fact, I consider what you call "underlaid long shots" as horses that require special attention (stable or "insider" betting, something I missed in handicapping, etc.) in every race I look at.

I guess the problem for me, as for every player, is how to identify true "underlays" or "overlays."

[Dave Schwartz]

I guess the problem for me, as for every player, is how to identify true "underlays" or "overlays."

That is a mouthful, Mike.

:ThmbUp:

[mountainman]

I'm as good as the next guy at detecting dead chalk-and very bold about denouncing such favorites on the air. But that step alone never ensures profit for me. Unless i have a reasonably strong opinion on a horse (or two) likely to capitalize, i'm better off just passing the race. Spreading has never worked for me. And that includes horizontal bets.

[teddy]

If you are good at finding dead chalk then the dutch bet becomes almost immediate winner if you are getting a fair win rebate. Finding things the public is not factoring is where I start. IE the 10 post at a mile at Mountaineer.

[Ocala Mike]

Now you're talking about what I can more relate to. In other words (obviously), removing underlaid CHALK is going to be far more profitable than removing underlaid LONG SHOTS. Also, if you're on to any kind of bias (speed/closer, post position, inside/outside) that the public is not on to (yet), you're way ahead of the game.

I remember when the rail going a mile out of the chute at the Big A was the kiss of death, and when on certain windy days at Belmont, outside closers (the "balcony move") won almost every race. Not sure if those biases still exist today, but pretty sure the public is "on" to them, if they do. Harder and harder to get an edge.

[DeltaLover]

Practically speaking, a large underlay creates value in all other starters. When and if this is the case you do not need to dutch in order to show a profit.

Randomly selecting a single starter among the overlays will show a positive return given a large enough sample (assuming of course that our underlay hypothesis is correct!)

The concept of dutching is overestimated. It does not add any value in your bottom line.

Even if we were able to estimate the 'real' odds line and base our strategy in over - underlays, optimal bankroll growth would suggest selecting the largest overlay instead of dutching all of them. I think this is an easy enough concept to grasp and prove.

More than this there is not sufficient evidence (setting aside proof!) that this 'real' line can be calculated or if it even exists.

[thaskalos]

I'm as good as the next guy at detecting dead chalk-and very bold about denouncing such favorites on the air. But that step alone never ensures profit for me. Unless i have a reasonably strong opinion on a horse (or two) likely to capitalize, i'm better off just passing the race. Spreading has never worked for me. And that includes horizontal bets.

Couldn't agree more! :ThmbUp:

I read a Dick Mitchell book a long time ago...in which he suggested that the horseplayer had a guaranteed edge on the game, if only he could effectively rule the favorite out as a win contender in the race.

"It doesn't matter WHOM you wager on"...Mitchell said..."stick to races with non-winning favorites...and you have a built-in edge on the game."

It sounded great to me, and I had a knack for finding losing favorites...so I followed his advice enthusiastically.

All the favorites lost...but none lost more than I...

[teddy]

Post 10, 11 , 12 returned .67 roi per $1 wagered last year. Expected roi should be .80 avg... so there is some advantage by removing all horses from post 10 out. Correct... or not...

[Capper Al]

That is a mouthful, Mike.

:ThmbUp:

Dave,

You are big on dutching, aren't you?

[Robert Fischer]

optimal bankroll growth would suggest selecting the largest overlay instead of dutching all of them. I think this is an easy enough concept to grasp and prove.

:ThmbUp:

This is correct when both


the bankroll is large enough to support low% win wagers , and,
a single "largest overlay" can be identified


When either of the above conditions can not be met, And a dutch play offers acceptable total value and total hit%, then you would opt to dutch.

[Dave Schwartz]

I am a fan of dutching.

Races fall into one of several categories:

1. Races where you have eliminated one or more low odds (3/1 and below) horses.

Note definition of "eliminate" is that the horse has (in your opinion) a much reduced chance of winning such that it would return its backers a huge net loss. This (effectively) fuels the profit on the other horses that you would consider wagering.

2. Races where 1 or more mid-priced horses (7/2 to 6/1) are "eliminated" such that the sum of the pool eliminated fuels the profit for the other horses that you would consider wagering.

3. Races where the top 1 or 2 public choice(s) have such an edge over the remaining horses that all the other horses can effectively be eliminated, fueling the profit on the the low priced horse(s).

4. Races where all the stars are lined up against you correctly such that there is an insignificant percentage of the pool that can be eliminated and no profitable wagers can be found.


Most horse players believe that #4 is the most common scenario. I find that not to be the case. I find that #2 is most common, followed by #1 and #3, respectively. Scenario #4 is the least common.

However, it all depends upon the individual player's tools and outlook.


Regards,
Dave Schwartz

[DeltaLover]

I am a fan of dutching.

Races fall into one of several categories:

1. Races where you have eliminated one or more low odds (3/1 and below) horses.

Note definition of "eliminate" is that the horse has (in your opinion) a much reduced chance of winning such that it would return its backers a huge net loss. This (effectively) fuels the profit on the other horses that you would consider wagering.

2. Races where 1 or more mid-priced horses (7/2 to 6/1) are "eliminated" such that the sum of the pool eliminated fuels the profit for the other horses that you would consider wagering.

3. Races where the top 1 or 2 public choice(s) have such an edge over the remaining horses that all the other horses can effectively be eliminated, fueling the profit on the the low priced horse(s).

4. Races where all the stars are lined up against you correctly such that there is an insignificant percentage of the pool that can be eliminated and no profitable wagers can be found.


Most horse players believe that #4 is the most common scenario. I find that not to be the case. I find that #2 is most common, followed by #1 and #3, respectively. Scenario #4 is the least common.

However, it all depends upon the individual player's tools and outlook.


Regards,
Dave Schwartz


Anyone can be a fan of anything, there is nothing wrong about it. There is people who are Yankee fans and also there is people who are Mets fans. In the other hand though there are things that we can hardly be fans of. For example it is very hard for someone to be a fan of a flat earth in our days.

The concept of 'dutching' based in a probability distribution that is matched against public's odds falls to the second category. Is a statement that can be approached analytically leaving no space for doubts.

As I stated previously I am a 'fan' of the idea that creating this probability distribution and use it as described is impossible or at least there is no proof that it can be done but this is another topic that we can discuss separately.

For the purpose of this conversation let's assume that this custom line indeed exists.

In this case case dutching two or more starters will almost never be the optimal approach.

Based in this line we will have an over - under factor for each horse indicating what is our percentage advantage - disadvantage for a bet to this horse.

The best bet we can make under this condition depends in the following two factors:

1) The overlay magnitute: meaning what is this percentage advantage presented by this starter

2) The absolute value of the odds provided by the pool. This is used because the possibility exists that although starter A might be a larger overlay than B, B will allow a larger bet without destroying the advantage. For example A and B might both offer 3% overlay but A is 25-1 while B is 5-1. If our bet will be $500 we better bet B since betting A will destroy the advantage.

Based in this the only case were we might dutch will be if both horses are exactly at the same odds, presenting exactly the same overlay value.

But, even in this case the final results in the 'long run' will be exactly the same as if we had bet one of the two overlays. In other words ducthing (even in this extreme case) does not increase the expected value it only makes bankroll fluctuations smoother.

Another example of this concept is the following:

We are lucky enough to find a casino that spreads the following bet: In a double zero roulette the payout of both zero and double zero instead of 35-1 they are 40-1 presenting overlays. In the long run it will make no difference if we bet both for $50 or any of the two for $100 in every spin. Our advantage will remain exactly the same. In the other hand if it was offering zero at 40-1 and double zero at 39-1 although both are overlays we should always bet only the zero since it is a larger one.

[raybo]

Anyone can be a fan of anything, there is nothing wrong about it. There is people who are Yankee fans and also there is people who are Mets fans. In the other hand though there are things that we can hardly be fans of. For example it is very hard for someone to be a fan of a flat earth in our days.

The concept of 'dutching' based in a probability distribution that is matched against public's odds falls to the second category. Is a statement that can be approached analytically leaving no space for doubts.

As I stated previously I am a 'fan' of the idea that creating this probability distribution and use it as described is impossible or at least there is no proof that it can be done but this is another topic that we can discuss separately.

For the purpose of this conversation let's assume that this custom line indeed exists.

In this case case dutching two or more starters will almost never be the optimal approach.

Based in this line we will have an over - under factor for each horse indicating what is our percentage advantage - disadvantage for a bet to this horse.

The best bet we can make under this condition depends in the following two factors:

1) The overlay magnitute: meaning what is this percentage advantage presented by this starter

2) The absolute value of the odds provided by the pool. This is used because the possibility exists that although starter A might be a larger overlay than B, B will allow a larger bet without destroying the advantage. For example A and B might both offer 3% overlay but A is 25-1 while B is 5-1. If our bet will be $500 we better bet B since betting A will destroy the advantage.

Based in this the only case were we might dutch will be if both horses are exactly at the same odds, presenting exactly the same overlay value.

But, even in this case the final results in the 'long run' will be exactly the same as if we had bet one of the two overlays. In other words ducthing (even in this extreme case) does not increase the expected value it only makes bankroll fluctuations smoother.

Another example of this concept is the following:

We are lucky enough to find a casino that spreads the following bet: In a double zero roulette the payout of both zero and double zero instead of 35-1 they are 40-1 presenting overlays. In the long run it will make no difference if we bet both for $50 or any of the two for $100 in every spin. Our advantage will remain exactly the same. In the other hand if it was offering zero at 40-1 and double zero at 39-1 although both are overlays we should always bet only the zero since it is a larger one.

Once again, someone posting that something is impossible ro a certain method is subservient to another one. This is horse racing, anything can happen, and many methods can be profitable. Just because a poster's own research says something doesn't work well, doesn't mean that it won't ever work or won't work for someone else.

Math and statistics, if they were the only way, would have made this game beatable by way more players than are currently beating it, because math and statistics people have been trying to beat this game forever, and yet there are very few winning players, still.

[Dave Schwartz]

In this case case dutching two or more starters will almost never be the optimal approach.

Thank you so much for clearing that up for me.

Now that I have this from a reliable source, I will give this nonsense up completely.

[DeltaLover]

Thank you so much for clearing that up for me.

Now that I have this from a reliable source, I will give this nonsense up completely.

I am reluctant to accept 'reliable sources' or argumentum ad verecundiam (http://en.wikipedia.org/wiki/Argument_from_authority) as a proof for anything, especially when this can be quantified.

I would expect a more constructive answer than this.

[raybo]

I think Delta is missing the point of "dutching" multiple horses. In the scenario he gave, with 1 horse at 5/1 and another at 25/1, he says you'd better bet the $500 on the 5/1 because your $500 bet would destroy the odds on the 25/1 horse. That is true if you were betting a flat bet on both horses. In dutching, you are not betting a flat bet on all your contenders, the lower priced horse will always receive a larger bet amount than the higher priced horse, so the odds on the 25/1 horse would not necessarily be destroyed by the dutching method, because less would be wagered on it, and more on the 5/1 horse.

As I said before, this is horse racing, many things during the running of the race can happen to destroy your perceived edge or advantage on a single contender. If that happens often enough, your perceived advantage is really not an advantage at all. By betting multiple horses, you are lessening the possibility of your advantage being destroyed by uncontrollable events occurring during the running of a race. What becomes a disadvantage for one contender may become a distinct advantage for one of your other contenders.

How often do uncontrollable events happen, that compromise the winning chances of the best horse, in the running of a race? Often enough that the vast majority of players making single selections are long term losers, and some of these losing players are some of the very best handicappers out there.

IMO, unless your determination of the actual winning probabilities of horses is extremely accurate (and that is probably extremely rare), betting a single contender is almost guaranteed to produce long term losses (of course, there are a few players who are exceptions, very very few).

[DeltaLover]

I think Delta is missing the point of "dutching" multiple horses. In the scenario he gave, with 1 horse at 5/1 and another at 25/1, he says you'd better bet the $500 on the 5/1 because your $500 bet would destroy the odds on the 25/1 horse. That is true if you were betting a flat bet on both horses. In dutching, you are not betting a flat bet on all your contenders, the lower priced horse will always receive a larger bet amount than the higher priced horse, so the odds on the 25/1 horse would not necessarily be destroyed by the dutching method, because less would be wagered on it, and more on the 5/1 horse.

As I said before, this is horse racing, many things during the running of the race can happen to destroy your perceived edge or advantage on a single contender. If that happens often enough, your perceived advantage is really not an advantage at all. By betting multiple horses, you are lessening the possibility of your advantage being destroyed by uncontrollable events occurring during the running of a race. What becomes a disadvantage for one contender may become a distinct advantage for one of your other contenders.

How often do uncontrollable events happen, that compromise the winning chances of the best horse, in the running of a race? Often enough that the vast majority of players making single selections are long term losers, and some of these losing players are some of the very best handicappers out there.

IMO, unless your determination of the actual winning probabilities of horses is extremely accurate (and that is probably extremely rare), betting a single contender is almost guaranteed to produce long term losses (of course, there are a few players who are exceptions, very very few).



The example of the 5-1 vs 25-1 has nothing to with dutching. It merely presents the importance of absolute odds as a factor in our decision when we have to select between two overlays.

The point I am trying to make is summarized as follows:

1) Dutching can only be correct only in very rare occasions namely when both the overlay magnitude and the odds are the same

2) Even in this rare case we gain nothing from dutching. In other words the final behavior of the strategy will be exactly the same

I gave the example with the roulette to make this effect more apparent.

If you have a different opinion please explain it by an example. Theories, stories and causation might be good to keep long conversations going but have really small value as predictors of future events.

[Dave Schwartz]

Delta,

You seem to have misunderstood my post. It IS my current methodology. I have been using it for quite awhile. And it IS profitable. For me.

Your somewhat dismissive pronouncement that it is without merit just goes to show how narrow any handicapper's thinking can be. Note that I include myself in that group.

We KNOW what we KNOW and have a natural tendency to assume that our way is the only way.

At least that is how your post sounded to me.


Back to the approach. The point of dutching is mitigate the consistency issue; to effectively monetizing profit from a good contender selection.

Of course, dutching the horses does not change the value of the individual horses. What it does do is change the definition of the bet so that the race becomes the entity instead of the horse.

Thus, I can offer a number such as "65% hit rate with average odds of 0.79:1."


I believe that the biggest issue the handicapper faces today is that "conventional value" is dead.


IOW, it is wonderful if you can build an engine that produces profitability against the odds. However, since there is no such thing as Final Odds to bet into, one must come up with a different approach.

I ran a test of around 3,000 races earlier this year. This test was done 1 race at a time, against the "final odds." At the 3,000 race mark I was profitable to the tune of almost 9% (before rebate). However, when I went to test live, the odds changes were so violent that I found I was making "incorrect wagers" in almost 75% of the races!

It literally turned me from a winner to a loser (not to mention making me more than a little crazy). After about 200 test plays I went back to the drawing board to find a better way to play. That was when my peeps and I came up with The Renegade Handicapper approach.

(Note: This is not a commercial. You can actually find the rudiments of the approach online here: Thanksgiving Improve Your Game Seminar (http://thehorsehandicappingauthority.com/thanksgiving-2012-improve-your-game-seminar/)

That was when I began to study tote movement. (Please do not confuse that with "charting.")

What I studied was the direction of the "final odds" from 0 minutes to post. Admittedly, the study was short because I was easily convinced early on that my data was accurate.

What I found was that at high-rebate tracks (such as MNR, CT, TUP), the ultimate winner was bet down in about 80% of the races. Conversely, the losers were generally bet upwards. (By "bet down" I mean their odds went lower.)

This was in stark contrast to tracks with very low rebates (such as California). At those tracks the typical movement was 40% down, 40% up and 20% unchanged.

Another interesting point was the percentage volume of handle that shows up in the pool after 0 minutes to post: Low rebate = 38% High rebate = 73%.


Use this info as you will. Meanwhile, I continue to use my Renegade Handicapper approach.


Regards,
Dave Schwartz

[thaskalos]

The example of the 5-1 vs 25-1 has nothing to with dutching. It merely presents the importance of absolute odds as a factor in our decision when we have to select between two overlays.

The point I am trying to make is summarized as follows:

1) Dutching can only be correct only in very rare occasions namely when both the overlay magnitude and the odds are the same

2) Even in this rare case we gain nothing from dutching. In other words the final behavior of the strategy will be exactly the same

I gave the example with the roulette to make this effect more apparent.

If you have a different opinion please explain it by an example. Theories, stories and causation might be good to keep long conversations going but have really small value as predictors of future events.

I have a slightly different view here.

You say that "we gain nothing from dutching".

People say the same about reversing exactas; it supposedly isn't the optimal way to make the bet. Others spend much time trying to determine what the optimal way is to handle their money while gambling.

There is a much bigger question that begs to be answered...IMO:

What is the player's risk tolerance level...and how does that relate to the betting methods that he has adopted?

If the player's risk tolerance level is high...then these "optimal" ways of playing may prove best.

If, on the other hand, his risk tolerance level is lower...then the optimal methods of playing are not for him...regardless of their added long-term appeal. This sort of player needs the added confidence that only additional hedging methods can provide.

The trick is not to follow the "optimal" method...but to find the method that suits our temperament and disposition.

[Jeff P]

Anyone can be a fan of anything, there is nothing wrong about it. There is people who are Yankee fans and also there is people who are Mets fans. In the other hand though there are things that we can hardly be fans of. For example it is very hard for someone to be a fan of a flat earth in our days.

The concept of 'dutching' based in a probability distribution that is matched against public's odds falls to the second category. Is a statement that can be approached analytically leaving no space for doubts.

As I stated previously I am a 'fan' of the idea that creating this probability distribution and use it as described is impossible or at least there is no proof that it can be done but this is another topic that we can discuss separately.

For the purpose of this conversation let's assume that this custom line indeed exists.

In this case case dutching two or more starters will almost never be the optimal approach.

Based in this line we will have an over - under factor for each horse indicating what is our percentage advantage - disadvantage for a bet to this horse.

The best bet we can make under this condition depends in the following two factors:

1) The overlay magnitute: meaning what is this percentage advantage presented by this starter

2) The absolute value of the odds provided by the pool. This is used because the possibility exists that although starter A might be a larger overlay than B, B will allow a larger bet without destroying the advantage. For example A and B might both offer 3% overlay but A is 25-1 while B is 5-1. If our bet will be $500 we better bet B since betting A will destroy the advantage.

Based in this the only case were we might dutch will be if both horses are exactly at the same odds, presenting exactly the same overlay value.

But, even in this case the final results in the 'long run' will be exactly the same as if we had bet one of the two overlays. In other words ducthing (even in this extreme case) does not increase the expected value it only makes bankroll fluctuations smoother.

Another example of this concept is the following:

We are lucky enough to find a casino that spreads the following bet: In a double zero roulette the payout of both zero and double zero instead of 35-1 they are 40-1 presenting overlays. In the long run it will make no difference if we bet both for $50 or any of the two for $100 in every spin. Our advantage will remain exactly the same. In the other hand if it was offering zero at 40-1 and double zero at 39-1 although both are overlays we should always bet only the zero since it is a larger one.

If in fact the player has a true mathematical edge, and provided the player can make enough plays to invoke the law of large numbers, the following formula determines the player's long term expectation:

P/L = (Handle) x (Edge)

I'll use your roulette example to illustrate. Let's assume for the sake of argument that the house is unaware that the 40-1 odds offered on 0 and the 39-1 offered on 00 generate a mathematical edge for the player.

Doing the math myself, I come up with a long term player edge of 10.8101% at 40-1 and 8.1081% at 39-1.

Let's further assume that the house limit at the roulette tables offering those odds is $1000.00 per spin and that over the course of several weeks a team of players manages to bet the max bet allowed by the house on both 0 and 00 for 100,000 spins.

For the situation where 40-1 is offered on 0, here's how that "should" work out for the team:

P/L = (Handle) x (Edge)

or

P/L = (100,000 x $1,000) x (.1081081)

P/L = ($100,000,000.00) x (.1081081)

P/L = $10,810,810.00


For the situation where 39-1 is offered on 00, here's how that "should" work out for the team:

P/L = (Handle) x (Edge)

or

P/L = (100,000 x $1,000) x (.081081)

P/L = ($100,000,000.00) x (.081081)

P/L = $8,108,100.00


Given the choice of betting into two situations that both offer long term profits, it should be obvious that betting into the situation with the greater edge will outperform betting into the situation with the lesser edge over time.

However, there's more to it than just that.

In the roulette example, provided the team of players has enough capital to bet the $1,000 house limit on both 0 and 00 without subjecting their pooled bankroll to undue risk (a subject for another discussion) the team will generate a higher P/L in the long run by betting both 0 and 00 - as opposed to betting 0 only.

Also, in the roulette example, the size of the player's bet doesn't negatively impact the odds.

In horse racing, the max that can be bet by the player without negatively impacting the payoff is a function of pool size. In other words, the player can bet more on a 5-1 shot than he can on a 25-1 shot in the same pool without impacting the payoff.



-jp

.

[DeltaLover]

If in fact the player has a true mathematical edge, and provided the player can make enough plays to invoke the law of large numbers, the following formula determines the player's long term expectation:

P/L = (Handle) x (Edge)

I'll use your roulette example to illustrate. Let's assume for the sake of argument that the house is unaware that the 40-1 odds offered on 0 and the 39-1 offered on 00 generate a mathematical edge for the player.

Doing the math myself, I come up with a long term player edge of 10.8101% at 40-1 and 8.1081% at 39-1.

Let's further assume that the house limit at the roulette tables offering those odds is $1000.00 per spin and that over the course of several weeks a team of players manages to bet the max bet allowed by the house on both 0 and 00 for 100,000 spins.

For the situation where 40-1 is offered on 0, here's how that "should" work out for the team:

P/L = (Handle) x (Edge)

or

P/L = (100,000 x $1,000) x (.1081081)

P/L = ($100,000,000.00) x (.1081081)

P/L = $10,810,810.00


For the situation where 39-1 is offered on 00, here's how that "should" work out for the team:

P/L = (Handle) x (Edge)

or

P/L = (100,000 x $1,000) x (.081081)

P/L = ($100,000,000.00) x (.081081)

P/L = $8,108,100.00


Given the choice of betting into two situations that both offer long term profits, it should be obvious that betting into the situation with the greater edge will outperform betting into the situation with the lesser edge over time.

However, there's more to it than just that.

In the roulette example, provided the team of players has enough capital to bet the $1,000 house limit on both 0 and 00 without subjecting their pooled bankroll to undue risk (a subject for another discussion) the team will generate a higher P/L in the long run by betting both 0 and 00 - as opposed to betting 0 only.

Also, in the roulette example, the size of the player's bet doesn't negatively impact the odds.

In horse racing, the max that can be bet by the player without negatively impacting the payoff is a function of pool size. In other words, the player can bet more on a 5-1 shot than he can on a 25-1 shot in the same pool without impacting the payoff.



-jp

.


I completely agree. Note though that this apporach is not dutching.

Dutching two overlays which are not equal will always result to a composite that will have less expectation that the higher one.

Of course having upper limits (either hard defined roulete or imposed by the pool size) might suggest a better approach to be a multi bet strategy which essentialy will treat each bet as part of a separate strategy.

[Dave Schwartz]

I completely agree. Note though that this apporach is not dutching.

Dutching two overlays which are not equal will always result to a composite that will have less expectation that the higher one.


And I completely agree with you.

However, this presupposes that your estimate of the horses' respective chances is accurate.

By combining horses together, this improves the likelihood that my probability estimate is reasonably accurate.

[castaway01]

I just wanted to say that I have found this thread very interesting, especially the information about the late money and odds direction of the winning horses at low-rebate and high-rebate tracks. That is excellent research.

[Magister Ludi]

Given: two horses in the same race, 5:1 and 25:1, with the same expected 3% overlay

BR = bankroll = $100,000
p = expected win probability
q = expected loss probability
A = net odds
e = (A+1)p-1>0
f* = fraction of BR to invest = e/A
f*BR = amount invested
EV = expected value = ef*BR

Optimal bet on horse a (5:1) only:

p(a) = .1717
A(a) = 5
e(a) = (5+1)*.1717-1 = .03
f*(a) = .03/5 = .006
f*(a)BR = $100,000*.006 = $600
EV(a) = $600*.03 = $18

Optimal bet on horse b (25:1) only:

p(b) = .0396
A(b) = 25
e(b) = (25+1)*.0396-1 = .03
f*(b) = .03/25 = .0012
f*(b)BR = $100,000*.006 = $120
EV(b) = $600*.03 = $3.6

Optimal hedge bets on both horses:

r = reserve rate = (1-sum(each p bet))/(1-sum(each 1/(A+1)))

r = (1-(.1717+.0396))/(1-((1/(5+1))+(1/(25+1))))) = .9923

f* = p-r/A
f*(a) = .1717-(.9923/6) = .00629
f*(b) = .0396-(.9923/26) = .00145

f*(a)BR = .00629*100,000 = $629
f*(b)BR = .00145*100,000 = $145

EV(a) = $629*.03 = $18.87
EV(b) = $145*.03 = $ 4.36
EV(total) = $23.23

EV(total)>EV(a)>EV(b)

[raybo]

And I completely agree with you.

However, this presupposes that your estimate of the horses' respective chances is accurate.

By combining horses together, this improves the likelihood that my probability estimate is reasonably accurate.

I thought I was inferring that in my post, regarding Delta's admonishment of multiple wagering . He says now that he was not referring to dutching. Ok, so he must have been referring to betting the same amount on both contenders. Jeff correctly identified the problem with that; your estimation of probability MUST be accurate, otherwise your perceived edge, and your perceived optimal betting method, go right out the window. One's ability to calculate accurate probabilities, in the first place, is highly improbable, due to the myriad of factors involved in horse racing, unlike the much more statistically "fixed" set of factors in a game of roulette. Trying to apply probabilities to both games, equally, is not only error filled, but downright insane, IMO.

By selecting a single selection in such an endeavor as horse racing, where we are at the mercy of so many negative possibilities involved in the running of races, one is therefore, at the mercy of that myriad of negative possibilities, with only one way out, that being, that single selection wins the race. In a method where one selects more than one horse, one's exposure to those same negative possibilities is reduced, considerably, and now has more than one way out, that being that any of the selections win. After all, the only time one realizes profit in a specific race, is when he has a bet on the actual winning horse, and that payout is large enough to produce profit in that race. Dutching allows that, as does a minimum odds requirement based on overall hit rate of the combined betting interests.

I also agree with Thas' post that some players need additional buffers against risk, due to their individual psychological makeup. While some have no problem with discipline and patience, being able to wait until their time comes to wager and risking large wagers on a single betting interest, most players cannot perform in such a manner, the risk/reward ratio is too high for their psyche to handle. Betting more favorable risk/reward ratios alleviates much of that, because one's hit rate is higher and the big, single event, losses are reduced greatly, allowing the player to keep his composure and consistency intact and to continue making sound decisions.

[Robert Fischer]

like i tried to convey in post 12:

If you have an edge and a bankroll, dutching is optimal when you have a block of contenders that can't be separated into a single best wager. Many times for example you will have a vulnerable heavy favorite, and two equal contenders from the same odds spectrum. Most of the time(in general, not when you have two equal contenders from the same odds spectrum) you will be able to identify a single best play, in which case a single play (not dutching) will be optimal.

The vast majority of players however, lack a positive expectation and/or a bankroll. These players can benefit from dutching in more situations (than the thriving pro from the first paragraph). These recreational or beginning players can also benefit from dutching even if they struggle with consistency and are occasionally profitable. Key benefits include protection of bankroll(from higher hit%), and protection from errors in separating contenders.

[DeltaLover]

By combining horses together, this improves the likelihood that my probability estimate is reasonably accurate.


Truth.

In this case though we deviate from the hypothesis of a precise probability distribution for each starter and we introduce the slightly different approach which is based in our belief that there is some inefficiency in the market but we are not sure exactly how it is distributed among the starters.

Most likely we can narrow it down to a subset of starters who seem to be the beneficiaries of the betting anomaly but we cannot be more granular.

Now, my point is that as long as we can reach this level of separation it irrelative how we are going to distribute our bet. My claim is that in this particular case dutching, betting only one starter selected from the overlaid group or betting them any possible amount (assuming of course we do not exceed pool's tolerance) all will have the same impact to the final behavior of our bankroll.

Most gamblers seem to not realize it, been tricked by the short term results thus jumping to betting schemes that have guaranteed inferiority over others that probably introduce higher fluctuations.

A common example is place vs win betting (especially in middle prices). Due to breakage it is easy to detect that although lower in striking rate the win approach is superior to place. Despite this there are quite a few gamblers who after their 5-2 shot finishes second they cannot resist the temptation to condemn their serf about not betting it for 'second' exactly because they are blinded by the immediate outcome.

If we are not prepared to loose much more bets than we are going to win and view each individual bet as a low significance component of a much larger process we cannot make really good decisions forming long term success.

[raybo]

Is it not accepted, by all, that a "precise probability distribution for each starter" is not possible? Or, is there someone out there who has such a precise probability distribution method? If the latter, let him speak up, otherwise, the former, for our purposes, is undeniable truth.

If the latter, then one is still not sure if his selection offers value because one does not know what the final odds will be until after one's bet is placed. Placing such a bet, on a single selection, simply because one believes value is present on that selection, is strictly speaking, a fantasy.

On the other hand, if the former is true, and it most certainly is, then one must find a way to make the probability distribution more precise, in order to find "true value". What other way is there other than: waiting for your single selection to offer odds that clearly exceed it's winning probability (which you cannot be precise about in the first place), or, selecting multiple contenders, which multiplies the winning probability of a single selection.

Is it really that hard to accept that "precise probability distribution for each starter" is, in reality, not possible? Do you know which horse will win the race, even with all the work and research you have put into your probabilities assignments? Be truthful, the answer is no.

[DeltaLover]

I don't think anyone claimed that it is even remotely possible to be exact to the estimate of the probability. Going this direction is comparable to chasing rainbows.

[Dave Schwartz]

In this case though we deviate from the hypothesis of a precise probability distribution for each starter and we introduce the slightly different approach which is based in our belief that there is some inefficiency in the market but we are not sure exactly how it is distributed among the starters.

The above would be me and, I believe, just about every other horse player on the planet.

The fact that one can proudly and loudly proclaim, "This is a 27.3% horse," does not make it accurate. (Yes, I know that you did not say it did.)

My Opinion
In order to be successful, each handicapper must find a way to maximize what he does best and minimize what he does poorly. (It helps if you can identify these problems first. :rolleyes: )

What my software and my approach does well is pick contenders. By comparison, its ability to distinguish between contenders is very difficult. I would hallucinate that this is a common problem for players.

Problem #1
If my contender picking is excellent and my final selection process not so excellent, how do I monetize what I have?


Problem #2
The last minute odds changes kill me. How do I overcome that, and, better yet, profit from it?


Solution (for me)
The solution for me... (and all who have an interest should pay close attention here, because this IS the essence of the Renegade Handicapper system)...

The solution for me was to:
1) Use the tote board to make the probabilities. (What?)

2) Penalize the non-contenders a huge part of their awarded probabilities and redistribute those probabilities to the contenders (i.e. normalize back to 100%).

3) Use those probabilities to bet into a properly adjusted morning line. (Huh?)

#1
A horse that has 30% of the pool gets a 30% chance of winning while a horse with 2% of the pool gets a 2% chance of winning. (Then make an adjustment for normal low-odds out-performing.)

#2
A horse that has a 30% chance of winning and is a non-contender gets penalized a huge amount of his hit rate. Let's say 90%, so he winds up with a 3% chance of winning. This isn't "correct" but it is as good as anything else. More on that a little later.

#3
Finally, I wager into the morning line.


Summary
My goal is:
1) To pick good contenders.

2) To have the public help me to understand which of my contenders are likely to be the best wagers. (i.e. most profitable)

3) If I dutch the better bets, then I wind up wagering according to ADJUSTED morning line. (That is, a line that fits the takeout in this race.)

If one dutches against the 0-min odds, most of the time they will wind up under-betting the winner and over-betting the loser(s) because 80% of the time the winner gets bet down.

If, instead, I bet against the M/L, a constant, what will happen is the bet will not change on the winner. This is an improvement over always betting less than should have been wagered.

Sure sounds like a crackpot idea, doesn't it?

Yet I am seeing it work.

[DeltaLover]

In order to be successful, each handicapper must find a way to maximize what he does best and minimize what he does poorly. (It helps if you can identify these problems first


The problem is that before the handicap finds the way to maximize what he does best he first has to find what is this that he is doing best. What are the criteria and what is the process needed for this?

Using a reductive approach we can conclude that there are two fundamental questions to be answered for any race:

- Should we bet or pass

- If we are going to bet what is the horse(s) we will use


A betting model (could be implemented as a computer program or it can be some manual process this is a small detail) should be able to answer both questions.

If it is not able to suggest exactly what the bet is going to look like but instead it gives you a set of contenders then it simply is not a complete.

This addresses your problem #1 very well..


By comparison, its ability to distinguish between contenders is very difficult.


You simply need more R&D until you have a more complete model. This is the area where an algorithmic approach can be proven most useful than anything else: to discover classifications where the empirical handicapper will believe that all starters look the same...

As far a problem #2 goes, again this can be part of the model. There is nothing superficial with the possible direction of late money. We can create models that are able to overcome this drift.

As far as your goal, I think that the real goal is to maximize bankroll growth. Picking good contenders or dutching better bets clearly is not the goal. Could be part of the means used to arrive to the goal.

[Dave Schwartz]

Delta Lover,

LOL - I find it interesting that instead of commenting on my SOLUTION you told me what I should do instead.

Let me make it perfectly clear. This is working for me.

You do not need to embrace it as your own. Neither do you need to agree with it.

However, I really do not need your suggestions to go back to doing what everyone else thinks you should do but doesn't seem to work.

Is what you are suggesting - basically, build a better line - working for you?

If it is working for you, please feel free to describe it further in layman's terms. Explain it to me like I am a 5th grader because I really want to understand.

I have laid out precisely what I do. Explain what you would suggest I do.

Please note that there is no sarcasm intended in this post.

Regards,
Dave Schwartz

[raybo]

The problem is that before the handicap finds the way to maximize what he does best he first has to find what is this that he is doing best. What are the criteria and what is the process needed for this?

Using a reductive approach we can conclude that there are two fundamental questions to be answered for any race:

- Should we bet or pass

- If we are going to bet what is the horse(s) we will use


A betting model (could be implemented as a computer program or it can be some manual process this is a small detail) should be able to answer both questions.

If it is not able to suggest exactly what the bet is going to look like but instead it gives you a set of contenders then it simply is not a complete.

This addresses your problem #1 very well..



You simply need more R&D until you have a more complete model. This is the area where an algorithmic approach can be proven most useful than anything else: to discover classifications where the empirical handicapper will believe that all starters look the same...

As far a problem #2 goes, again this can be part of the model. There is nothing superficial with the possible direction of late money. We can create models that are able to overcome this drift.

As far as your goal, I think that the real goal is to maximize bankroll growth. Picking good contenders or dutching better bets clearly is not the goal. Could be part of the means used to arrive to the goal.

So, you are saying, I think(? whew), that any method that includes betting your contenders, instead of a single selection, is incomplete? If so, then I guess my method is incomplete and I should throw it in the trash, even though it produces net profit.

Or, are you saying that such a method needs to be further refined, in some way? If so, should that further refinement be to "maximize bankroll growth"? Isn't that what everyone who already has a successful method tries to do by keeping the method current and updated, doing more research, and continuing to implement improvements?

Or, are you saying that you have "the way" to make anyone's method complete? If so, please enlighten us. I assume this comes with a written guarantee, and can not fail?

[DeltaLover]

Is what you are suggesting - basically, build a better line - working for you?



Certainly I am not suggesting building a better line.


Explain it to me like I am a 5th grader because I really want to understand.


I did not claim that I have the solution. I have just defined the conditions.

Let me repeat:

I am suggesting a model that provides very concrete selections, not allowing any room for human tweaking and intervention.

Based in what you described this is not the case with your approach.

You are stating that:


By comparison, its ability to distinguish between contenders is very difficult


I see this as a weakness of a 'near by' solution .

You do not clearly describe how you eventually are reaching your final selection. Is this selection deterministic or is open to every individual interpretations?

[Dave Schwartz]

Certainly I am not suggesting building a better line.

Then I do not understand what you are suggesting. That seemed like a logical guess.



My final selections are based upon:

1. Using the probabilities outlined in the previous post and multiplying by the adjusted morning line odds' payout.

2. This results in a $net and optimum bet.

3. This gives me the ability to decide which horses are "better" bets.

IOW, the higher optimum bets are the best (theoretical) bets.

[dnlgfnk]

Dave,

I suspect I may have stumbled onto a similar approach such as you've described. The single most valuable piece of information I have received from this site is the table of odds, number of starters and winning pct. at each odds increment, as achieved by the public. I believe it was Jeff P. who produced the table last year or so.

I had always heard that the public were the best handicappers race after race, but now I saw proof of it, thanks to Jeff. It was remarkable to me that the public even correctly identified 6th ranked choices over 7th, 7th over 8th, etc. over many races.

The next step seemed obvious to me. Working from the M/L highest odds= lowest pct. chance of winning horses, one can combine the pcts. of the higher odds horses until they match the chances of the favorite(s). Then by eliminating certain members of the higher odds group using subjective factors, as well as certain favorite(s), one can often zero in on accurate overlays.

I too work off the M/L at the Cal tracks, with a bit of tinkering in the case of a questionable individual posted price, which isn't often.

[teddy]

Couldn't agree more! :ThmbUp:

I read a Dick Mitchell book a long time ago...in which he suggested that the horseplayer had a guaranteed edge on the game, if only he could effectively rule the favorite out as a win contender in the race.

"It doesn't matter WHOM you wager on"...Mitchell said..."stick to races with non-winning favorites...and you have a built-in edge on the game."

It sounded great to me, and I had a knack for finding losing favorites...so I followed his advice enthusiastically.

All the favorites lost...but none lost more than I...

Im intrigued that someone as respected as you could be capable of losing after selecting losing favorites. Does that not insure you are a winner dutching the rest of the field. In most cases. Were you not dutching the rest of the field. I know you must get at least 5 pts on the win rebates.

[MightBeSosa]

Im intrigued that someone as respected as you could be capable of losing after selecting losing favorites. Does that not insure you are a winner dutching the rest of the field. In most cases. Were you not dutching the rest of the field. I know you must get at least 5 pts on the win rebates.

I think the problem is the word 'effectively', which does not mean 100%. Eliminating losing 1-1 shots at a high rate is not an easy task, and unless you're world class, you will be wrong plenty.

That's why most players try to get a better return than a simple dutch, as they know they probably can't make any money that way.

[Capper Al]

I have been succesful using a combination of oddsline and elimination rules while waiting for miminum odds to wager. This means that my top rated horse with my lowest odds will be passed by even with high odds on the tote-board if the elimination rules kick it out.

[atlasaxis]

Couldn't agree more! :ThmbUp:

I read a Dick Mitchell book a long time ago...in which he suggested that the horseplayer had a guaranteed edge on the game, if only he could effectively rule the favorite out as a win contender in the race.

"It doesn't matter WHOM you wager on"...Mitchell said..."stick to races with non-winning favorites...and you have a built-in edge on the game."

It sounded great to me, and I had a knack for finding losing favorites...so I followed his advice enthusiastically.

All the favorites lost...but none lost more than I...


Which book is that out of? Thanks.

[DeltaLover]

So, you are saying, I think(? whew), that any method that includes betting your contenders, instead of a single selection, is incomplete? and can not f


Any 'method' that is open to human interpretations and interactions to form its final selections is incomplete. Instead it should provide concrete selections that should become the only bets we place in the race. No human interaction is permitted.

I really do not understand how you arrived to the single selection. Surely, we can more than one selections. My point is that dutching is not increasing our expected value at no case.


If so, then I guess my method is incomplete and I should throw it in the trash, even though it produces net profit.


I dont recall mentioning about thawing any method to trash. When you say it produces net profit, can we back test your claim it using a database? Do you have a program that I can run can to back test it so I can confirm what you are claiming? Do you keep sufficient records for the performance of your system? How big of a bankroll do I need to assure a profit of lets's say 30K / year? What is the maximum draw down I should expect assuming a fixed bet per race?

Since you continually refrain to the same argument that your system produces net profit, let me ask you a simple question: How much is your average bet and what is your total handle per year? It makes a big difference if you bet recreationally in the range of several dollars per bet than betting professionally amounts that can change the payout of the race.



Or, are you saying that you have "the way" to make anyone's method complete? If so, please enlighten us. I assume this comes with a written guarantee, and can not fail?


I would never trumpet something like this and if you go through my postings you will never see me claiming that I am showing a net profit, I beat the game etc.

In the hypothetical case that someone has such a model I see absolutely no reason for him to share it with anyone making it a commercial application.

If from anyone a written guarantee should be asked, it is from someone promoting a horse betting system or method that shows a net profit, don't you think?

[raybo]

Any 'method' that is open to human interpretations and interactions to form its final selections is incomplete. Instead it should provide concrete selections that should become the only bets we place in the race. No human interaction is permitted.

I really do not understand how you arrived to the single selection. Surely, we can more than one selections. My point is that dutching is not increasing our expected value at no case.



I dont recall mentioning about thawing any method to trash. When you say it produces net profit, can we back test your claim it using a database? Do you have a program that I can run can to back test it so I can confirm what you are claiming? Do you keep sufficient records for the performance of your system? How big of a bankroll do I need to assure a profit of lets's say 30K / year? What is the maximum draw down I should expect assuming a fixed bet per race?

Since you continually refrain to the same argument that your system produces net profit, let me ask you a simple question: How much is your average bet and what is your total handle per year? It makes a big difference if you bet recreationally in the range of several dollars per bet than betting professionally amounts that can change the payout of the race.





I would never trumpet something like this and if you go through my postings you will never see me claiming that I am showing a net profit, I beat the game etc.

In the hypothetical case that someone has such a model I see absolutely no reason for him to share it with anyone making it a commercial application.

If from anyone a written guarantee should be asked, it is from someone promoting a horse betting system or method that shows a net profit, don't you think?

That post was referencing your constant implcations that the best way to play the horses is via math and statistics. If you don't mean your posts to be construed as such then perhaps you should proofread them before submitting.

As to my claims, there is much documentation available, both here and other places on the web. I doubt you could follow the rules for profitable play, because you could never accept the simplicity of the method.

[DeltaLover]

Ray,

it all boils down to statistics whatever approach you follow.

As long as input and output are clearly defined, in our case input will be the historical data and output the betting selections, the hidden layers of processing that are forming the final decision can as simple of complex as possible, using advanced math or astrology, this can be seen as an implementation detail. What really counts is the performance of the model which will be measured by statistical methods.

Now if we want to become more specific and delve in to the guts of the implementation there are quite a few thinks that we might be interested of..

For example since you are referring to rules of play, some one can ask questions like the following:

Do you have a concrete example? Do your rules resemble a pure finite state machine or are they also has expert system characteristics?

Are they more inclined towards determinism or they present stochastic characteristics as well?

What is your data normalization strategy? How you are grouping metrics and transform them to comparable ratings? Do you use some ranking for data transformation? How you treat outliers? Do you make a distinction based in the fact if a metric is following the normal distribution or not?

The list goes on and on. Designing a successful handicapping 'software' is not as simple as it sounds. It requires a vast array of knowledge and experience to design and implement it successfully and some who is qualified for the task is unlikely to deal with it since he can apply his skills to other more profitable ventures.

The other interesting thing that I detect is that those who have formed some handicapping method or software seem to be very stubborn about their approach and tend to ignore what other have to say.

Take for example this thread:

http://www.paceadvantage.com/forum/showthread.php?p=1367739&#post1367739

which ends with a very provocative statement from one of the most knowledgeable guys among the posters in this board.


Nobody even cared to respond. To ask for a clarification, even to challenge...

Someone is telling us that he has model with a 64% striking rate and 1.16 ROI then silence, the thread ends.... Very strange!

[Dave Schwartz]

Someone is telling us that he has model with a 64% striking rate and 1.16 ROI then silence, the thread ends.... Very strange!

There are two reasons for that, IMHO.

1. The majority of posters do not possess enough math skills to understand it and CANNOT ask intelligent questions.

2. Those that have the math skills either understood what he said or (more likely) have their own rabbit trail they are chasing.


Besides, when a guy says "I am getting this wonderful result" they logical result of asking, "Could you please explain EXACTLY how you did that so that I can duplicate it?" is likely to be answered with a succinct "NO!"

I suggest that you are in the #2 category yourself.

[raybo]

Ray,

it all boils down to statistics whatever approach you follow.

As long as input and output are clearly defined, in our case input will be the historical data and output the betting selections, the hidden layers of processing that are forming the final decision can as simple of complex as possible, using advanced math or astrology, this can be seen as an implementation detail. What really counts is the performance of the model which will be measured by statistical methods.

Now if we want to become more specific and delve in to the guts of the implementation there are quite a few thinks that we might be interested of..

For example since you are referring to rules of play, some one can ask questions like the following:

Do you have a concrete example? Do your rules resemble a pure finite state machine or are they also has expert system characteristics?

Are they more inclined towards determinism or they present stochastic characteristics as well?

What is your data normalization strategy? How you are grouping metrics and transform them to comparable ratings? Do you use some ranking for data transformation? How you treat outliers? Do you make a distinction based in the fact if a metric is following the normal distribution or not?

The list goes on and on. Designing a successful handicapping 'software' is not as simple as it sounds. It requires a vast array of knowledge and experience to design and implement it successfully and some who is qualified for the task is unlikely to deal with it since he can apply his skills to other more profitable ventures.

The other interesting thing that I detect is that those who have formed some handicapping method or software seem to be very stubborn about their approach and tend to ignore what other have to say.

Take for example this thread:

http://www.paceadvantage.com/forum/showthread.php?p=1367739&#post1367739

which ends with a very provocative statement from one of the most knowledgeable guys among the posters in this board.


Nobody even cared to respond. To ask for a clarification, even to challenge...

Someone is telling us that he has model with a 64% striking rate and 1.16 ROI then silence, the thread ends.... Very strange!

My point, regarding your posts, is that you seem to always revert back to creating a piece of "software" and then insist that everything in that software be objective in nature, "As long as input and output are clearly defined, in our case input will be the historical data and output the betting selections,", etc..

Then you further point out that each player's software must conform to statistical confirmation, etc..

There are many players who do not create "software", nor use objective data, nor use statistical confirmation in order to determine if their approach is viable. I assume you would consider all those players to be losing players and their methods "incomplete".

As I stated earlier, and many here have also, there are many ways to be successful in horse racing, and one only needs to look at his bankroll to determine if one's method is viable. You obviously, according to your posts, think that is archaic, and totally unreliable, and therefore, unbelievable. You must, at some point, agree that you have no idea if it is unreliable or not.

The reason people ignore your postings, and others, like the one you linked to by Mike, is because of the insistence that everything boils down to math, statistics, and software. While, for those people, your posts, and others', do not apply to them, and are overly technical and appear pompous, in nature, to them.

Mike's posting was probably ignored for the same reason, and that's probably why there were no responses to it.

Yes, he, like you and others here, may be extremely educated/intellectual, and extremely knowledgeable of horse racing, in general, but unless those views/information/beliefs, etc, are posted in such a way as to not alienate people who don't understand, or have an interest (as well as the necessary formal education to comprehend and implement such methods) in such approaches, then those people will turn you, and those others, off, out of hand.

Posting examples of your methodology, and it's results, accomplishes much more than just stating higher educational teachings and beliefs, that may or may not be applicable to horse racing, and/or monetary success or failure, regarding horse racing. For example, using math/statistical means to attempt to relate a casino game like black jack or roulette to horse racing, is like relating apples to oranges, yes they are both fruits, but the differences between the 2 are huge, and not comparable.

[raybo]

Someone is telling us that he has model with a 64% striking rate and 1.16 ROI then silence, the thread ends.... Very strange!

Why would that seem strange to you? Is it because you consider that unbelievable? Or are you thinking that some sort of negative response should have been posted? Or, someone asking for the key to his success?

Why didn't you respond to it? Because you automatically disbelieved it? If the latter, why? Were you waiting for him to prove it to you, just out of the goodness of his heart? If so, why don't you prove that your approach works? Or why don't you just post some examples of the results of your approach being successful, long term.

It's one thing for someone to suggest his/her approach/methodology works and post some examples of that long term success. It's quite another, for someone to continually suggest an approach/methodology, while never posting any examples of that method actually working, long term.

[DeltaLover]

Presenting a complete winning approach is something different than discussing the approach, the components and the research direction to be taken. Again I do not claim I have a winning approach so based in this I have nothing to show. In the other hand, I have quite a few of silent evidence, of things that do not work which is equally important.

What I can discuss in depth is the reduction of data, their transformations and the objectives of the process.

Clarification of these fundamental topics is the more important objective and I see very little effort spend towards this direction.

To give you a better understanding let me tell you that although I have absolutely no interest in any application that receives the past performance files and presents their contents in a different way, even applying some simple statistics, like IV, ROI etc I would have been very keen about a provider of a specific set a metrics (like prime power for instance) without even knowing their name, what they measure or how they are generated, assuming I find it significant for modeling purposes. This is a black box that I can successfully adopt and use. How I will use it is something that I have interest not to reveal it unless I no longer care about its use. Tor verification though, I can provide you with a higher level, metric which will have more predictive value than the original one.

[ceejay]

If, instead, I bet against the M/L, a constant, what will happen is the bet will not change on the winner. This is an improvement over always betting less than should have been wagered.

So, considering that the morning line typically over prices favorites you end up under weighting favorites who are contenders. Right? (I know, I need to buy "renegade")

[raybo]

So, considering that the morning line typically over prices favorites you end up under weighting favorites who are contenders. Right? (I know, I need to buy "renegade")

By "M/L", I believe he is referring to his program's "line", not the morning line produced by the track handicapper.

[Robert Fischer]

oops

[Dave Schwartz]

So, considering that the morning line typically over prices favorites you end up under weighting favorites who are contenders. Right? (I know, I need to buy "renegade")

No. In fact, more likely the opposite.

I see a lot of favorites as good bets; as underbet.

BTW, there is a sale going up tonight where you can get a significant discount on just about everything.

[teddy]

BLACK FRIDAY NO DOUBT.

[teddy]

I dutched 2 horses at aqueduct today to reduce my risk of race mishaps. One broke down in front of the other. Both wiped out... So much for that idea.

[raybo]

I dutched 2 horses at aqueduct today to reduce my risk of race mishaps. One broke down in front of the other. Both wiped out... So much for that idea.

Well, you could have bet all that cash on one, or the other, of those horses, and still lost the same amount. It's just another lost horse race, of many more to come.

[Magister Ludi]

I dutched 2 horses at aqueduct today to reduce my risk of race mishaps. One broke down in front of the other. Both wiped out... So much for that idea.
You should always bet on ALL of the overlays in a race. See #26.

[teddy]

Im am trying to play that on paper, bettting all the overlays. So far not so good... Considering leaving out the 30-1 and up

Playing only races where the chalk is a considerable underlay.. IE my line says 3-1 and it is going off at even money.

[Magister Ludi]

Im am trying to play that on paper, bettting all the overlays. So far not so good... Considering leaving out the 30-1 and up
Optimally betting all of the overlays is how Cetacea swill so much krill.

[traynor]

My point, regarding your posts, is that you seem to always revert back to creating a piece of "software" and then insist that everything in that software be objective in nature, "As long as input and output are clearly defined, in our case input will be the historical data and output the betting selections,", etc..

Then you further point out that each player's software must conform to statistical confirmation, etc..

There are many players who do not create "software", nor use objective data, nor use statistical confirmation in order to determine if their approach is viable. I assume you would consider all those players to be losing players and their methods "incomplete".

As I stated earlier, and many here have also, there are many ways to be successful in horse racing, and one only needs to look at his bankroll to determine if one's method is viable. You obviously, according to your posts, think that is archaic, and totally unreliable, and therefore, unbelievable. You must, at some point, agree that you have no idea if it is unreliable or not.

The reason people ignore your postings, and others, like the one you linked to by Mike, is because of the insistence that everything boils down to math, statistics, and software. While, for those people, your posts, and others', do not apply to them, and are overly technical and appear pompous, in nature, to them.

Mike's posting was probably ignored for the same reason, and that's probably why there were no responses to it.

Yes, he, like you and others here, may be extremely educated/intellectual, and extremely knowledgeable of horse racing, in general, but unless those views/information/beliefs, etc, are posted in such a way as to not alienate people who don't understand, or have an interest (as well as the necessary formal education to comprehend and implement such methods) in such approaches, then those people will turn you, and those others, off, out of hand.

Posting examples of your methodology, and it's results, accomplishes much more than just stating higher educational teachings and beliefs, that may or may not be applicable to horse racing, and/or monetary success or failure, regarding horse racing. For example, using math/statistical means to attempt to relate a casino game like black jack or roulette to horse racing, is like relating apples to oranges, yes they are both fruits, but the differences between the 2 are huge, and not comparable.

Anyone who uses a series of past events--of whatever size--as the basis for prediction of future events is using (or misusing) "statistics." That is not rocket science, and no PhD is required. It is simply a fact that pooh-poohing statistics as irrelevant does absolutely nothing to change. If your application uses past races to determine "pace scenarios" for betting on future races, it uses statistics. If it did not use statistics, there would be no purpose whatsoever in using past races.

Whether your application uses statistics competently or correctly is another matter. Or are your "pace scenarios" so highly developed that one can simply feed in details of today's races, and out pops a generous ROI? Without the intermediate step of using some kind of simplistic "statistical analysis" of past races to determine a "pace scenario"? I think that is what DeltaLover is trying to say in regard to an "incomplete application."

It is silly to continually criticize something that your application apparently depends on. "Brohamer models" are not new, nor are they rocket science. They even work occasionally. For awhile. In some cases.

[teddy]

During my test of betting all the overlays when the race has a vulnerable favorite, by my definition. Should I test using dutching or just flat $2 bet on all horses that are overlays to the same line. Doing this I know for a fact that most 30-1 and up horses are horrible wagers. Many times you could get twice the odds on a long shot on ehorsex exchange years ago.

[Magister Ludi]

During my test of betting all the overlays when the race has a vulnerable favorite, by my definition. Should I test using dutching or just flat $2 bet on all horses that are overlays to the same line. Doing this I know for a fact that most 30-1 and up horses are horrible wagers. Many times you could get twice the odds on a long shot on ehorsex exchange years ago.

If you are sure that they are all overlays, you should size your bets in accordance with the Kelly Criterion. Here you can learn how to do that: http://www.kentuckyderby.org/betting/hedging/.

If you're not sure, use this betting progression: Initialize the sequence with a bet of 0 units. Decrease your bet by 1 unit after each win, loss or push. This system has the highest expectation and lowest variance of any progression ever devised for random horse betting. Apologies to James Grosjean.

[traveler]

Then I do not understand what you are suggesting. That seemed like a logical guess.



My final selections are based upon:

1. Using the probabilities outlined in the previous post and multiplying by the adjusted morning line odds' payout.

2. This results in a $net and optimum bet.

3. This gives me the ability to decide which horses are "better" bets.

IOW, the higher optimum bets are the best (theoretical) bets.


I'm confused how they can be the best "theoretical" bets, they either are or they aren't. I know some will say we never have a large enough sample size to prove anything, but can you say - "I tested 100 races and the higher optimum bets WERE the best bets"?

[raybo]

Anyone who uses a series of past events--of whatever size--as the basis for prediction of future events is using (or misusing) "statistics." That is not rocket science, and no PhD is required. It is simply a fact that pooh-poohing statistics as irrelevant does absolutely nothing to change. If your application uses past races to determine "pace scenarios" for betting on future races, it uses statistics. If it did not use statistics, there would be no purpose whatsoever in using past races.

Whether your application uses statistics competently or correctly is another matter. Or are your "pace scenarios" so highly developed that one can simply feed in details of today's races, and out pops a generous ROI? Without the intermediate step of using some kind of simplistic "statistical analysis" of past races to determine a "pace scenario"? I think that is what DeltaLover is trying to say in regard to an "incomplete application."

It is silly to continually criticize something that your application apparently depends on. "Brohamer models" are not new, nor are they rocket science. They even work occasionally. For awhile. In some cases.

Of course my method uses statistics. I wasn't referring to my method, I was referring to those players who don't use statistics, in their true calculated form, in their play or their research. Yes, they are probably using statistics accumulated inside their brain from past experiences, but not in a true calculated manner. Some keep records and some don't. That doesn't mean that all of them are losing players. And the sooner the math/statistics experts here realize that, the sooner long drawn out discussions, like this, about these matters will decline in frequency, and we can get back to horse racing.

[raybo]

Im am trying to play that on paper, bettting all the overlays. So far not so good... Considering leaving out the 30-1 and up

Playing only races where the chalk is a considerable underlay.. IE my line says 3-1 and it is going off at even money.

Maybe your line is at fault?

[teddy]

Maybe your line is at fault?
yah, its mechanical, not subjective to any personal handicapping. Trying to get a starting point that makes sense..

Races where the favorite is considered legit, then the longs shots perform very poorly. But I am not looking for legit chalks.

[traynor]

Of course my method uses statistics. I wasn't referring to my method, I was referring to those players who don't use statistics, in their true calculated form, in their play or their research. Yes, they are probably using statistics accumulated inside their brain from past experiences, but not in a true calculated manner. Some keep records and some don't. That doesn't mean that all of them are losing players. And the sooner the math/statistics experts here realize that, the sooner long drawn out discussions, like this, about these matters will decline in frequency, and we can get back to horse racing.

I never left horse racing. My mistake--I thought you were saying that statistics could not possibly be useful in horse race analysis--which is, of course, absurd. I think the more one understands statistics and data modeling, the better race analyst he or she becomes.

[Dave Schwartz]

I'm confused how they can be the best "theoretical" bets,

LOL - The key word here is "best" - I suppose I should have said theoretical "best bets" instead.

[raybo]

I never left horse racing. My mistake--I thought you were saying that statistics could not possibly be useful in horse race analysis--which is, of course, absurd. I think the more one understands statistics and data modeling, the better race analyst he or she becomes.

Everyone is free to think what they want. That doesn't necessarily mean that one's method is better than another's. The bottom line is the bottom line, regardless of how you get there.

[traynor]

Everyone is free to think what they want. That doesn't necessarily mean that one's method is better than another's. The bottom line is the bottom line, regardless of how you get there.

I agree. That is why I don't accept anyone else's claims about results. I don't really care whether he or she or someone's brother-in-law in Poughkeepsie is (or at least claims to be) making money hand over fist. I am primarily interested in my own results, and that is the only real "proof" I accept as valid.

That is why I write my own race analysis software. That is also why I use standard statistical analysis and data modeling methodologies in those applications to assure that I am taking full advantage of the available data.

[thaskalos]

I agree. That is why I don't accept anyone else's claims about results. I don't really care whether he or she or someone's brother-in-law in Poughkeepsie is (or at least claims to be) making money hand over fist. I am primarily interested in my own results, and that is the only real "proof" I accept as valid.

That is why I write my own race analysis software. That is also why I use standard statistical analysis and data modeling methodologies in those applications to assure that I am taking full advantage of the available data.

I think that this is it, in a nutshell.

NONE of us accept anybody else's lofty claims...and we are all disinterested and suspicious of any "proof" that others provide as validation for their results in this game.

So...where do we go from here?

[eurocapper]

Statistics can be good when one has all the data. But I don't think past performances represent all there is to know about horse racing (like race replays or trainer familiarity). What's worse, it's info that is public and everyone is using.

[atlasaxis]

Couldn't agree more! :ThmbUp:

I read a Dick Mitchell book a long time ago...in which he suggested that the horseplayer had a guaranteed edge on the game, if only he could effectively rule the favorite out as a win contender in the race.

"It doesn't matter WHOM you wager on"...Mitchell said..."stick to races with non-winning favorites...and you have a built-in edge on the game."

It sounded great to me, and I had a knack for finding losing favorites...so I followed his advice enthusiastically.

All the favorites lost...but none lost more than I...

Which book is that out of? Thanks.

[raybo]

I think that this is it, in a nutshell.

NONE of us accept anybody else's lofty claims...and we are all disinterested and suspicious of any "proof" that others provide as validation for their results in this game.

So...where do we go from here?

Posted plays/results/screenshots are as good as it gets. If those aren't acceptable then anyone wanting to improve their play will have to do it the hard way, usually meaning years of hard work and frustration, guaranteeing nothing, of course.

[DeltaLover]

Can you please provide a link to the plays?

[raybo]

Can you please provide a link to the plays?

"Search"

[DeltaLover]

thx for the quick and enlighting answer! very helpfull...

[raybo]

thx for the quick and enlighting answer! very helpfull...

So you want me to do the search for you? Are you spelling disabled? Or just wanting something for nothing? Nothing worth anything is free in this world.

[teddy]

Posted plays/results/screenshots are as good as it gets. If those aren't acceptable then anyone wanting to improve their play will have to do it the hard way, usually meaning years of hard work and frustration, guaranteeing nothing, of course.


Those are not acceptable.. I found that out long ago.


One of the guys said, you only need to look at bankroll size to determine if your successful. I think that is pretty much simplifying the whole argument you guys have on here as to who is right.

[TrifectaMike]

So you want me to do the search for you? Are you spelling disabled? Or just wanting something for nothing? Nothing worth anything is free in this world.


So, Mr. Raybo, how much is ALL DATA Free (http://www.paceadvantage.com/forum/showthread.php?goto=newpost&t=65749) worth? Absolutely NOTHING according to you!

Mike (Dr Beav)

[raybo]

So, Mr. Raybo, how much is ALL DATA Free (http://www.paceadvantage.com/forum/showthread.php?goto=newpost&t=65749) worth? Absolutely NOTHING according to you!

Mike (Dr Beav)

You must have a version of Excel on your computer and, of course, you have to purchase the data files. So, even AllData free workbooks aren't free.

By the way, I don't keep a folder of play links I have posted here and other places, so I, too, would have to do a search to find them. If you or Delta are too lazy to look them up, then you guys have a problem I can't fix for you.

[raybo]

Those are not acceptable.. I found that out long ago.

Then you're on your own.


One of the guys said, you only need to look at bankroll size to determine if your successful. I think that is pretty much simplifying the whole argument you guys have on here as to who is right.

That would be me.

[DeltaLover]

By the way, I don't keep a folder of play links I have posted here and other places, so I, too, would have to do a search to find them. If you or Delta are too lazy to look them up, then you guys have a problem I can't fix for you.

I am really surprised to hear this..

I would expect a utility that can back test the results of the model in a matter of a seconds. Without something like this I really cannot see how anyone can have any confidence in his research.

So, in the case you want to change something in your model, how do you test its behavior?

Can you please describe your approach?


As far as been lazy, I have to admit it... This link has something to say about this:

http://blogoscoped.com/archive/2005-08-24-n14.html

[Magister Ludi]

yah, its mechanical, not subjective to any personal handicapping. Trying to get a starting point that makes sense..

Races where the favorite is considered legit, then the longs shots perform very poorly. But I am not looking for legit chalks.

Many seem to think that you need to "add value" by handicapping. Not so. You can ferret out inefficiencies in public information and be very successful simply by observing it, recognizing it for what it is, and acting upon it. You are on the right track. I will say no more about this.

[raybo]

I am really surprised to hear this..

I would expect a utility that can back test the results of the model in a matter of a seconds. Without something like this I really cannot see how anyone can have any confidence in his research.

So, in the case you want to change something in your model, how do you test its behavior?

Can you please describe your approach?


As far as been lazy, I have to admit it... This link has something to say about this:

http://blogoscoped.com/archive/2005-08-24-n14.html

Yes, it's called "auto-record" mode, with the click of a single button the program imports all the cards and results you have selected from your 2 data file folders, using the batch processing method in the program, processes each race, in turn, compares the program's picks with the results, records the wins, losses, payouts, track, distance, surface, PPG, black box selection method, etc., deletes the oldest file from the enclosed database, then adds the one just processed and recorded, in that database. It then proceeds, as above, until the last card and results are imported, processed, and recorded. All played races, as determined by automatic pass/play notifications within the program, are ready for viewing, as well as a summary of the completed process, showing hit rate, ROI, and net profit, for all minimum odds ranges from 1/9 or higher through 11/1 or higher. This minimum odds ranges summary also shows each of the 3 possible individual picks' hit rate, ROI, and net profit, for each of the minimum odds ranges listed above.

[Robert Fischer]

The whole idea that a player's method has to be 100% objective, and include no human element of subjectivity is a joke.

This goes for the business of horseplaying, as well as the business of offering handicapping products.

The only time a method must be 100% objective is in the business of testing methods against a database.

[DeltaLover]

The only time a method must be 100% objective is in the business of testing methods against a database.

This is exactly what I am saying

[Ocala Mike]

Optimally betting all of the overlays is how Cetacea swill so much krill.



Good one! Like netting tuna commercially, you really can't care about a few unfortunate dolphins that you drown.

:rolleyes:

[Dave Schwartz]

The whole idea that a player's method has to be 100% objective, and include no human element of subjectivity is a joke.

I completely disagree.

[DeltaLover]

I completely disagree.

You mean that you disagree that you can not objectively back test a dessision process that cannot be quantified rather relying in intuition and personal judgement instead?

Frankly, I cannot see how something like this can be achived...

Can you please explain ?

[traynor]

I think that this is it, in a nutshell.

NONE of us accept anybody else's lofty claims...and we are all disinterested and suspicious of any "proof" that others provide as validation for their results in this game.

So...where do we go from here?

Generalities. Specifics. Theoretical underpinnings. Observations. Personal opinions. Brainstorming. New views on old topics. Personal insights. Shared information. The potential is unlimited.

What forums are supposed to be.

[raybo]

The whole idea that a player's method has to be 100% objective, and include no human element of subjectivity is a joke.

This goes for the business of horseplaying, as well as the business of offering handicapping products.

The only time a method must be 100% objective is in the business of testing methods against a database.

My program is completely objective, whether in "daily play" mode or while testing/researching, against the database, in "auto-record" mode. No user input is required once the user's settings preferences are selected in the program.

[traynor]

Yes, it's called "auto-record" mode, with the click of a single button the program imports all the cards and results you have selected from your 2 data file folders, using the batch processing method in the program, processes each race, in turn, compares the program's picks with the results, records the wins, losses, payouts, track, distance, surface, PPG, black box selection method, etc., deletes the oldest file from the enclosed database, then adds the one just processed and recorded, in that database. It then proceeds, as above, until the last card and results are imported, processed, and recorded. All played races, as determined by automatic pass/play notifications within the program, are ready for viewing, as well as a summary of the completed process, showing hit rate, ROI, and net profit, for all minimum odds ranges from 1/9 or higher through 11/1 or higher. This minimum odds ranges summary also shows each of the 3 possible individual picks' hit rate, ROI, and net profit, for each of the minimum odds ranges listed above.

That sounds a lot like backfitting. For those unfamiliar with the coding process, backfitting is the continual tweaking of a selection process so that the application makes selections based on what would have worked best in the group of races studied if the tweaking had been done before the races were run.

[raybo]

That sounds a lot like backfitting. For those unfamiliar with the coding process, backfitting is the continual tweaking of a selection process so that the application makes selections based on what would have worked best in the group of races studied if the tweaking had been done before the races were run.

Backfitting requires the results to be known prior to the testing. That does not happen in my program. A track's first weeks of cards and results are used to populate the "initial" database with applicable data (one can also use the same time period from the previous meet as the initial database). Once that initial database has been populated, the testing/daily play begins, grabbing the applicable data, from the initial database, and used for selections in each newly processed race. The results are not known for the race being processed, of course. Once that new card has been processed, the oldest card is removed from the initial database and the one that was just processed is added to the database, keeping it dynamic regarding time of year and time of meet.

Any "tweaking" that is done, by the user, is done before new cards are imported and processed. And once those new settings are made to the program's method and minimum odds settings, they are not reset until, and if, the user wishes to retest that track.

The summary report for each testing session can be sent to a separate worksheet, and retained there, so the user can check them going forward. These checks can be used to reveal a recent, more profitable minimum odds setting. Usually the only setting that may change, is the minimum odds setting, rarely does the rankings method change during a meet. At Delta so far this year, I started with 5/2 or higher minimum odds, as shown in the summary report as the best odds range, and recently retested this year's cards leading to my changing that minimum odds setting to 3/1 or higher, due to an increase in hit rate, ROI, and net profit at those odds. The previous 5/2 odds were still quite profitable, but 3/1 had become even more profitable than it was earlier in the meet, so I changed that setting, based on those recent dynamic changes in the database. 3/1 or higher was the setting I used in Delta's last meet, and this year is pointing to that setting being the best one again this year. The summary report shows the combination of both the earlier setting and the new setting. It does not recalculate the earlier race settings, once recorded they remain recorded and reflected in the summary report. So, my summary report shows all the races played at 5/2 odds and also the recent races played at 3/1 odds.

[Dave Schwartz]

You mean that you disagree that you can not objectively back test a dessision process that cannot be quantified rather relying in intuition and personal judgement instead?

Frankly, I cannot see how something like this can be achived...

Can you please explain ?

I am a firm believer that the system must be exactly that: completely systematic. That means no user-defined input whatsoever.



I am not sure where the communication breakdown here is. I thought that was clear.

[DeltaLover]

Backfitting requires the results to be known prior to the testing. That does not happen in my program. A track's first weeks of cards and results are used to populate the "initial" database with applicable data (one can also use the same time period from the previous meet as the initial database). Once that initial database has been populated, the testing/daily play begins, grabbing the applicable data, from the initial database, and used for selections in each newly processed race. The results are not known for the race being processed, of course. Once that new card has been processed, the oldest card is removed from the initial database and the one that was just processed is added to the database, keeping it dynamic regarding time of year and time of meet.

Any "tweaking" that is done, by the user, is done before new cards are imported and processed. And once those new settings are made to the program's method and minimum odds settings, they are not reset until, and if, the user wishes to retest that track.

The summary report for each testing session can be sent to a separate worksheet, and retained there, so the user can check them going forward. These checks can be used to reveal a recent, more profitable minimum odds setting. Usually the only setting that may change, is the minimum odds setting, rarely does the rankings method change during a meet. At Delta so far this year, I started with 5/2 or higher minimum odds, as shown in the summary report as the best odds range, and recently retested this year's cards leading to my changing that minimum odds setting to 3/1 or higher, due to an increase in hit rate, ROI, and net profit at those odds. The previous 5/2 odds were still quite profitable, but 3/1 had become even more profitable than it was earlier in the meet, so I changed that setting, based on those recent dynamic changes in the database. 3/1 or higher was the setting I used in Delta's last meet, and this year is pointing to that setting being the best one again this year. The summary report shows the combination of both the earlier setting and the new setting. It does not recalculate the earlier race settings, once recorded they remain recorded and reflected in the summary report. So, my summary report shows all the races played at 5/2 odds and also the recent races played at 3/1 odds.

Seems like a interesting approach.

Do you use any data coming from other tracks / meeting except what you describe here?

[DeltaLover]

I think this is exactly what Rob is saying here:

The only time a method must be 100% objective is in the business of testing methods against a database.

So, where is the disagreement?

[raybo]

Seems like a interesting approach.

Do you use any data coming from other tracks / meeting except what you describe here?

Only the data included in the raw Brisnet or JCapper/HDW data files, that is imported into the program.

The "initial" database, as I stated, can be populated with the same time period from the previous year's meet. So, if the user wants to play the first weeks of this year's meet (of course, this year's card and results files don't exist yet), he/she can use that same time period from last year's meet in the "initial" database.

[raybo]

I think this is exactly what Rob is saying here:



So, where is the disagreement?

That is true, unless you are talking about a "black box" in which case, the method must be 100% objective during daily play also. Or, it's not a true "black box".

[traynor]

Backfitting requires the results to be known prior to the testing. That does not happen in my program. A track's first weeks of cards and results are used to populate the "initial" database with applicable data (one can also use the same time period from the previous meet as the initial database). Once that initial database has been populated, the testing/daily play begins, grabbing the applicable data, from the initial database, and used for selections in each newly processed race. The results are not known for the race being processed, of course. Once that new card has been processed, the oldest card is removed from the initial database and the one that was just processed is added to the database, keeping it dynamic regarding time of year and time of meet.

Any "tweaking" that is done, by the user, is done before new cards are imported and processed. And once those new settings are made to the program's method and minimum odds settings, they are not reset until, and if, the user wishes to retest that track.

The summary report for each testing session can be sent to a separate worksheet, and retained there, so the user can check them going forward. These checks can be used to reveal a recent, more profitable minimum odds setting. Usually the only setting that may change, is the minimum odds setting, rarely does the rankings method change during a meet. At Delta so far this year, I started with 5/2 or higher minimum odds, as shown in the summary report as the best odds range, and recently retested this year's cards leading to my changing that minimum odds setting to 3/1 or higher, due to an increase in hit rate, ROI, and net profit at those odds. The previous 5/2 odds were still quite profitable, but 3/1 had become even more profitable than it was earlier in the meet, so I changed that setting, based on those recent dynamic changes in the database.

That is an interesting explanation. However, any modeling that uses past race results could be accurately defined as backfitting. That is the basis for most modeling processes--the problem is the degree to which individual and possibly aberrant events in that clump of races are replicated in the present and future. In short, how reliable is it to bet on today's races using what would have worked betting on yesterday's races if one had known it before yesterday's races were run?

That is not intended as a criticism of your approach, or of your application. It is a reality that just about everyone who bets on horse races has to deal with on a daily basis. It may be the way you have explained the process you use, but it seems based on events in the group of races used. That--in turn--suggests that it is only accurate to the degree that those events are replicated in current and future races. In short, it suffers the same deficiencies as most other modeling processes.

[raybo]

That is an interesting explanation. However, any modeling that uses past race results could be accurately defined as backfitting. That is the basis for most modeling processes--the problem is the degree to which individual and possibly aberrant events in that clump of races are replicated in the present and future. In short, how reliable is it to bet on today's races using what would have worked betting on yesterday's races if one had known it before yesterday's races were run?

That is not intended as a criticism of your approach, or of your application. It is a reality that just about everyone who bets on horse races has to deal with on a daily basis. It may be the way you have explained the process you use, but it seems based on events in the group of races used. That--in turn--suggests that it is only accurate to the degree that those events are replicated in current and future races. In short, it suffers the same deficiencies as most other modeling processes.

Unless there has been a huge change at the track you're playing since the testing was done, the results, going forward, are remarkably similar. The reason all recorded races are displayed in the summary section, is so the user can look for "outliers" that may skew the net results. The user can,of course, delete those "outliers" from that section, which will then be reflected in the new net results. The program, if used according to the instructions and wagering rules, hits horses that you might consider "outliers", however, the sheer number of similar paying hits tells me that only an extreme payout, or 2, are true "outliers". In other words the program picks both low priced winners and high priced winners, but we only place wagers on those picks that are at or above our minimum odds requirement, usually between 2/1 or higher and 4/1 or higher, depending on the testing from those individual tracks. Some tracks year in and year out, have higher average payouts, some have lower payouts, year in and year out. Our track testing shows that those tracks with average payouts that are above the average are the most profitable ones to play. Of course hit rate plays a part in that determination, and is clearly indicated in the summary report. I use a combination of; number of plays, hit rate, ROI, and net profit to determine the settings I will use for a particular track. I am fortunate in that I don't want or need "action", so number of plays is relatively unimportant to me. However, other players want, or need, that "action", this program allows them to tailor their settings to accomplish a larger number of plays, while still remaining profitable. Bankroll turn-over can accomplish good profit, even at lower net ROIs, so the action players aren't stuck losing money, simply because they want, or need, to play lots of races. That doesn't even include the use of rebate shops, which can enhance the monetary results.

[traynor]

Unless there has been a huge change at the track you're playing since the testing was done, the results, going forward, are remarkably similar. The reason all recorded races are displayed in the summary section, is so the user can look for "outliers" that may skew the net results. The user can,of course, delete those "outliers" from that section, which will then be reflected in the new net results. The program, if used according to the instructions and wagering rules, hits horses that you might consider "outliers", however, the sheer number of similar paying hits tells me that only an extreme payout, or 2, are true "outliers". In other words the program picks both low priced winners and high priced winners, but we only place wagers on those picks that are at or above our minimum odds requirement, usually between 2/1 or higher and 4/1 or higher, depending on the testing from those individual tracks. Some tracks year in and year out, have higher average payouts, some have lower payouts, year in and year out. Our track testing shows that those tracks with average payouts that are above the average are the most profitable ones to play. Of course hit rate plays a part in that determination, and is clearly indicated in the summary report. I use a combination of; number of plays, hit rate, ROI, and net profit to determine the settings I will use for a particular track. I am fortunate in that I don't want or need "action", so number of plays is relatively unimportant to me. However, other players want, or need, that "action", this program allows them to tailor their settings to accomplish a larger number of plays, while still remaining profitable. Bankroll turn-over can accomplish good profit, even at lower net ROIs, so the action players aren't stuck losing money, simply because they want, or need, to play lots of races. That doesn't even include the use of rebate shops, which can enhance the monetary results.

Something seems to be getting lost in translation, probably because of the way I have posed questions. I understand that your application models by displaying "what would have happened in this group of races if ..." Specifically, your application makes selections. Does the application alter the criteria used to make those selections based on the results of the races in the group used?

My reason for asking is that it seems unusual to me that an application using a month of races (for example) at a given track would use different criteria to make selections today than it would use a week from now (with an additional week of races added/deleted). That implies a massive study of the optimal dataset to use and you have made no mention of doing such. For example, how do you know that the results obtained by using X races (in the "last 30 days," "last 45 days," or whatever) would not be better or worse using some other frame?

I am not asking for empirical evidence. I am asking for the underlying logic. It seems--from your descriptions--that you are aiming from a moving position at a moving target. It is the updating part that I find confusing--it seems you are saying that (regardless of filter criteria--assume they are equal) the applications method of selection would be different today than it would be a week from now. Or that 30 days of past races would result in different selections being made by the application today than 35 days of past races.

[traynor]

If my questions seem unusual, it is probably because you have never used AllWays (or a similar software application) that "tweaks itself" (a euphemism for backfitting) based on specific groups of races.

[raybo]

Something seems to be getting lost in translation, probably because of the way I have posed questions. I understand that your application models by displaying "what would have happened in this group of races if ..." Specifically, your application makes selections. Does the application alter the criteria used to make those selections based on the results of the races in the group used?

My reason for asking is that it seems unusual to me that an application using a month of races (for example) at a given track would use different criteria to make selections today than it would use a week from now (with an additional week of races added/deleted). That implies a massive study of the optimal dataset to use and you have made no mention of doing such. For example, how do you know that the results obtained by using X races (in the "last 30 days," "last 45 days," or whatever) would not be better or worse using some other frame?

I am not asking for empirical evidence. I am asking for the underlying logic. It seems--from your descriptions--that you are aiming from a moving position at a moving target. It is the updating part that I find confusing--it seems you are saying that (regardless of filter criteria--assume they are equal) the applications method of selection would be different today than it would be a week from now. Or that 30 days of past races would result in different selections being made by the application today than 35 days of past races.

As I have stated many times in the past, the program uses running styles, early speed points, versus pace pressure to make some of its eliminations. It uses ranges of running styles and early speed points, that have won the majority of specific pace pressure categories. These ranges change as races are removed or added to the database. So yes, the criteria changes as the meet goes forward, reflecting the current state of the database, which, while remaining the same size, continually updates itself regarding running styles and early speed point ranges of recent winners.

This program, in a sense, learns from itself, but not as in a NN or AI method.

The reason our basic database size is 24-25 cards is due to the need to remain recent regarding time of year and time of meet. We want to stay as current as possible regarding horses running, jockey colony, trainer colony, weather conditions, especially general temperature and moisture ranges, etc.. We don't want to handicap moist, cold weather races using dry, warm weather races in the database, for example. 250 races, or so, works quite well for most tracks, but at other tracks the user can alter the number of cards/races he/she retains in the database, and find an optimal number that produces the best results, regarding what factor(s) they prefer (# of plays, hit rate, ROI, profit, or combinations of those).

The eliminations are just a portion of the program, the underlying portion is the velocity rankings, and those algorithms don't change, they simply rank the remaining contenders, from which we bet up to 3 of those top ranked contenders, if any, or all, of them meet our minimum odds requirements.

So, in a way, we are starting from a fixed point, progressing to a recent moving point, and shooting at a near term future moving target.

We think the game is constantly evolving, as the meet progresses. What won 3 or 4 weeks ago, may not win today, tomorrow, or the next few weeks. Due to changes in the overall environment at that particular track.

[traynor]

As I have stated many times in the past, the program uses running styles, early speed points, versus pace pressure to make some of its eliminations. It uses ranges of running styles and early speed points, that have won the majority of specific pace pressure categories. These ranges change as races are removed or added to the database. So yes, the criteria changes as the meet goes forward, reflecting the current state of the database, which, while remaining the same size, continually updates itself regarding running styles and early speed point ranges of recent winners.

This program, in a sense, learns from itself, but not as in a NN or AI method.

The reason our basic database size is 24-25 cards is due to the need to remain recent regarding time of year and time of meet. We want to stay as current as possible regarding horses running, jockey colony, trainer colony, weather conditions, especially general temperature and moisture ranges, etc.. We don't want to handicap moist, cold weather races using dry, warm weather races in the database, for example. 250 races, or so, works quite well for most tracks, but at other tracks the user can alter the number of cards/races he/she retains in the database, and find an optimal number that produces the best results, regarding what factor(s) they prefer (# of plays, hit rate, ROI, profit, or combinations of those).

The eliminations are just a portion of the program, the underlying portion is the velocity rankings, and those algorithms don't change, they simply rank the remaining contenders, from which we bet up to 3 of those top ranked contenders, if any, or all, of them meet our minimum odds requirements.

So, in a way, we are starting from a fixed point, progressing to a recent moving point, and shooting at a near term future moving target.

We think the game is constantly evolving, as the meet progresses. What won 3 or 4 weeks ago, may not win today, tomorrow, or the next few weeks. Due to changes in the overall environment at that particular track.

That is why I am reluctant to accept at face value (whatever that may be) figures that state how many winners and at what prices (for example, at Delta Downs) were selected over some period of time, because the process used to select those entries was continually changing. That is not a criticism, or an implication that the stats you provide links to are not accurate.

It is to point out that one of the major difficulties facing race analysis software developers is that each seemingly trivial tweak essentially makes previous models obsolete--winners selected at point-in-time A may have been excluded by the tweak if viewed from point-in-time B. In essence, if the selection process has changed between the two points, so have the results--selections made between point-in-time A and point-in-time B would very definitely NOT have been made using the criteria established at point-in-time B--they would have been selected using the criteria that existed at point-in-time A (and likely excluded the winners selected using the criteria adopted at point-in-time B).

It is not rocket science, nor is it a new problem. Race analysts (as well as most modelers and data analysts) have understood the difficulties almost from the start. It is a defect built-in to the "self-teaching software" paradigm, whatever it is called. What one ends up with at point-in-time whatever is just what he or she had at point-in-time A--a set of values that would have worked if the results of the races had been known before they were run. In short, backfitting to a specific group of races.

[raybo]

That is why I am reluctant to accept at face value (whatever that may be) figures that state how many winners and at what prices (for example, at Delta Downs) were selected over some period of time, because the process used to select those entries was continually changing. That is not a criticism, or an implication that the stats you provide links to are not accurate.

It is to point out that one of the major difficulties facing race analysis software developers is that each seemingly trivial tweak essentially makes previous models obsolete--winners selected at point-in-time A may have been excluded by the tweak if viewed from point-in-time B. In essence, if the selection process has changed between the two points, so have the results--selections made between point-in-time A and point-in-time B would very definitely NOT have been made using the criteria established at point-in-time B--they would have been selected using the criteria that existed at point-in-time A (and likely excluded the winners selected using the criteria adopted at point-in-time B).

It is not rocket science, nor is it a new problem. Race analysts (as well as most modelers and data analysts) have understood the difficulties almost from the start. It is a defect built-in to the "self-teaching software" paradigm, whatever it is called. What one ends up with at point-in-time whatever is just what he or she had at point-in-time A--a set of values that would have worked if the results of the races had been known before they were run. In short, backfitting to a specific group of races.

You're welcome to your opinion, of course. Mine is that if your method is static, it may work in a portion of a meet, but will not continue to work throughout the remainder of the meet, because things during every meet change, and the static selection method you use early in the meet, will likely not continue to work later in the meet.

THAT, IMO, is the fallacy of most handicapping methods. People expect nothing to change, and once, they think, that they have found "the key", they expect it will remain the key. That, IMO, is incorrect.

The "specific group of races" you mention, in my program, are not specific. Those races change every time you remove a card and add a new card to the database. We are playing races live, then they are added to the database, races that the database has not seen before. Yes, during testing you are using the the same set of races, only to get a starting point, so you have a better chance of playing the first races of the new meet on a positive note. As old cards are removed and new ones added, the database more efficiently and accurately reflects what is happening at the track, what types of horses are winning what types of races, and at what kinds of prices and hit rates.

But, not everyone believes this way, so they shouldn't use the program, because their lack of faith in it will allow them to deviate from the method and the rules. That will doom them to failure, for sure.

And, with that, I'm done explaining my program in this thread. So, save this explanation for the next time you want it explained again. It gets old and redundant.

[traynor]

You're welcome to your opinion, of course. Mine is that if your method is static, it may work in a portion of a meet, but will not continue to work throughout the remainder of the meet, because things during every meet change, and the static selection method you use early in the meet, will likely not continue to work later in the meet.

THAT, IMO, is the fallacy of most handicapping methods. People expect nothing to change, and once, they think, that they have found "the key", they expect it will remain the key. That, IMO, is incorrect.

The "specific group of races" you mention, in my program, are not specific. Those races change every time you remove a card and add a new card to the database. We are playing races live, then they are added to the database, races that the database has not seen before. Yes, during testing you are using the the same set of races, only to get a starting point, so you have a better chance of playing the first races of the new meet on a positive note. As old cards are removed and new ones added, the database more efficiently and accurately reflects what is happening at the track, what types of horses are winning what types of races, and at what kinds of prices and hit rates.

But, not everyone believes this way, so they shouldn't use the program, because their lack of faith in it will allow them to deviate from the method and the rules. That will doom them to failure, for sure.

And, with that, I'm done explaining my program in this thread. So, save this explanation for the next time you want it explained again. It gets old and redundant.

I think the thing that is doomed to failure is exactly what happened to a LOT of people who tried to use Sartin's methodology and the "Brohamer Models" that were an intrinsic part of it. Things change as a meet progresses--that is a basic fact of life. The belief that continually tweaking a selection process so that it would have selected yesterdays winners if you had had it yesterday somehow "adjusts to those changes" is a bit of foolishness best avoided.

[myhorse1]

:)

If you're not sure, use this betting progression: Initialize the sequence with a bet of 0 units. Decrease your bet by 1 unit after each win, loss or push. This system has the highest expectation and lowest variance of any progression ever devised for random horse betting. Apologies to James Grosjean.


initializing your bet with 0 units and going down from there seems to offer very little upside potential.

[Magister Ludi]

:)


initializing your bet with 0 units and going down from there seems to offer very little upside potential.

It was merely meant to be a humorous way of expressing that your optimal investment is 0 if you don't have a positive expected value.

[myhorse1]

It was merely meant to be a humorous way of expressing that your optimal investment is 0 if you don't have a positive expected value.

my response was also meant as humor.

thread drift ---sorry

[teddy]

Someone figure this up for me so I can decide it its worth a shot..

IF in a 12 horse field I have 3 non contenders to pitch that are each 25-1. Then how much does that add to my roi bottom line. 9%....this is based on the fact that they only win 1% of the time. SO I kn ow I will lose 1 bet of 100. When one runs in.

[MightBeSosa]

Someone figure this up for me so I can decide it its worth a shot..

IF in a 12 horse field I have 3 non contenders to pitch that are each 25-1. Then how much does that add to my roi bottom line. 9%....this is based on the fact that they only win 1% of the time. SO I kn ow I will lose 1 bet of 100. When one runs in.

3 at 25-1, same as a 9-1 shot, so you say you can be right about tossing a 9-1 shot 99 times out of a 100.

Congrats!

[teddy]

3 at 25-1, same as a 9-1 shot, so you say you can be right about tossing a 9-1 shot 99 times out of a 100.

Congrats!


Thats what my software says... 1% winners.

[Red Knave]

Thats what my software says... 1% winners.
I checked 12 horse fields of non-maiden sprinters for a period ending June 30, 2012. I came up with 4387 races. These are the results. Interestingly, horses at exactly 25-1 slightly outperformed. As an example, the line labelled "9" is the total of all horses in the sample from above 6.99 to 1 (i.e. 7-1) up tp 9.99 to 1.
25 is exactly 25.00 to 1 to 25.99 to 1.

[Red Knave]

25-1 and above win at 1.4% in this sample.

[sammy the sage]

Someone figure this up for me so I can decide it its worth a shot..

IF in a 12 horse field I have 3 non contenders to pitch that are each 25-1. Then how much does that add to my roi bottom line. 9%....this is based on the fact that they only win 1% of the time. SO I kn ow I will lose 1 bet of 100. When one runs in.

Seems to be MUCH higher% when I'm betting... :lol: :bang:

[OverlayHunter]

Red Knave -

I just want to be sure of what your sample is "saying" in the context of teddy's question.

My understanding is that he was asking about a 12 horse field with 3 of the 12 sitting at 25:1 and that your results indicate that each of those 25:1 horses - had they been in your sample - would have roughly a 1.4% chance to win and that the chance of one of the 3 winning would be 4.2% (1.4% x 3) and the probability of one of the other 9 horses winning would be 95.8%.

Is that correct?

[MightBeSosa]

25-1 and above win at 1.4% in this sample.

??

Your chart seems to bear out exactly what I said. 25-1 will win about 3% , give or take. Thats 9% for three of them.

Mixing in 99-1 shots is not useful in this context.

We have a claim here that they can get that down to 1% (from 9).

Right. :eek:

You'd be a very rich man booking on betfair.

[Red Knave]

Red Knave -

I just want to be sure of what your sample is "saying" in the context of teddy's question.

My understanding is that he was asking about a 12 horse field with 3 of the 12 sitting at 25:1 and that your results indicate that each of those 25:1 horses - had they been in your sample - would have roughly a 1.4% chance to win and that the chance of one of the 3 winning would be 4.2% (1.4% x 3) and the probability of one of the other 9 horses winning would be 95.8%.

Is that correct?
In this admittedly small sample, horses at exactly 25-1 won at 3.2%. I didn't know if the OP used 25-1 as a proxy for all horses going off at 25-1 or more so I added the post about that option also. Horses in this sample who went off at 25-1 or higher (so, yes, including 35-1, 99-1 etc) won at 1.4%.
So, if there were three 25-1 horses then I would put their chances at 3 * 3.2% or approximately what MightBeSosa says, 9-1.

[Capper Al]

3 at 25-1, same as a 9-1 shot, so you say you can be right about tossing a 9-1 shot 99 times out of a 100.

Congrats!

You're right, how about narrowing it down to a couple of low priced horses and a long shot. Hopefully by picking 3 or 4 horses out of 8 or more starters would make us reconsider.

[lansdale]

Hi Dave,

Considering the nature of some of my comments about you in the past, and some of the exchanges between us, you and PA readers can apply whatever degree of skepticism that seems appropriate.

That said, your comments here on tote-watching brought to mind the one phone converation I had with you. I had mentioned the idea that the two leading contenders, sometimes but not always favorites, seemed to 'split', close to post-time (BTW I think I mentioned 2 minutes, not 0), and the one whose odds then dropped further was winning more than 50% of the races. I was wondering whether this was worth pursuing. Although you seemed to be very much aware of this phenomenon you dismissed this idea as being too chalky, and I generally agreed. Also, my sample was very small. It seems as though you have returned to this concept, but refined both the applicable locations (high rebate tracks), and the time (less than zero minutes), and found a useful play.

I'm curious whether that, since these high-rebate tracks also seem to be smaller, cheaper tracks, inside knowledge is also something of a factor, as was recently suggested by Thaskolos, in comparing how inside knowledge should be weighted at smaller vs. bigger tracks.

BTW, in case it seems otherwise - not trying to claim any credit - this was just something I stumbled across.

I'd be interested in knowing the sample size, although I know that's a phrase that's verboten for some. Also would like to know average odds of 'dropping' winners and what % are favorites.

Cheers,

lansdale


Delta,

You seem to have misunderstood my post. It IS my current methodology. I have been using it for quite awhile. And it IS profitable. For me.

Your somewhat dismissive pronouncement that it is without merit just goes to show how narrow any handicapper's thinking can be. Note that I include myself in that group.

We KNOW what we KNOW and have a natural tendency to assume that our way is the only way.

At least that is how your post sounded to me.


Back to the approach. The point of dutching is mitigate the consistency issue; to effectively monetizing profit from a good contender selection.

Of course, dutching the horses does not change the value of the individual horses. What it does do is change the definition of the bet so that the race becomes the entity instead of the horse.

Thus, I can offer a number such as "65% hit rate with average odds of 0.79:1."


I believe that the biggest issue the handicapper faces today is that "conventional value" is dead.


IOW, it is wonderful if you can build an engine that produces profitability against the odds. However, since there is no such thing as Final Odds to bet into, one must come up with a different approach.

I ran a test of around 3,000 races earlier this year. This test was done 1 race at a time, against the "final odds." At the 3,000 race mark I was profitable to the tune of almost 9% (before rebate). However, when I went to test live, the odds changes were so violent that I found I was making "incorrect wagers" in almost 75% of the races!

It literally turned me from a winner to a loser (not to mention making me more than a little crazy). After about 200 test plays I went back to the drawing board to find a better way to play. That was when my peeps and I came up with The Renegade Handicapper approach.

(Note: This is not a commercial. You can actually find the rudiments of the approach online here: Thanksgiving Improve Your Game Seminar (http://thehorsehandicappingauthority.com/thanksgiving-2012-improve-your-game-seminar/)

That was when I began to study tote movement. (Please do not confuse that with "charting.")

What I studied was the direction of the "final odds" from 0 minutes to post. Admittedly, the study was short because I was easily convinced early on that my data was accurate.

What I found was that at high-rebate tracks (such as MNR, CT, TUP), the ultimate winner was bet down in about 80% of the races. Conversely, the losers were generally bet upwards. (By "bet down" I mean their odds went lower.)

This was in stark contrast to tracks with very low rebates (such as California). At those tracks the typical movement was 40% down, 40% up and 20% unchanged.

Another interesting point was the percentage volume of handle that shows up in the pool after 0 minutes to post: Low rebate = 38% High rebate = 73%.


Use this info as you will. Meanwhile, I continue to use my Renegade Handicapper approach.


Regards,
Dave Schwartz

[Dave Schwartz]

Lansdale,

Years ago (before the whale era) I studied the drop downs. At that point I did not find there was anything "meaningful" in the results. That has changed, as has racing.

As for the data itself, I think I will work on a more official study before commenting on it too much more. What I did was a cursory sample - by my usual standards closer to anecdotal than empirical - so there is room for improvement.

I will say this... it appears to me that the about 80% of the winners are bet down at high rebate tracks from the 1st flash after 0 minutes to post. In comparison, at low or none rebate tracks it is more like 40% up, 40% down, 20% unchanged.


Regards,
Dave Schwartz

[lansdale]

Thanks for the headsup Dave. I'll be interested to see any further notes on this idea. My best to you and yours this holiday season.

Cheers,

lansdale

Lansdale,

Years ago (before the whale era) I studied the drop downs. At that point I did not find there was anything "meaningful" in the results. That has changed, as has racing.

As for the data itself, I think I will work on a more official study before commenting on it too much more. What I did was a cursory sample - by my usual standards closer to anecdotal than empirical - so there is room for improvement.

I will say this... it appears to me that the about 80% of the winners are bet down at high rebate tracks from the 1st flash after 0 minutes to post. In comparison, at low or none rebate tracks it is more like 40% up, 40% down, 20% unchanged.


Regards,
Dave Schwartz

[barn32]

If you are sure that they are all overlays, you should size your bets in accordance with the Kelly Criterion. Here you can learn how to do that: http://www.kentuckyderby.org/betting/hedging/.
Nobody on God's green earth uses the Kelly Criterion to size their bets. Nobody.

[therussmeister]

Nobody on God's green earth uses the Kelly Criterion to size their bets. Nobody.
I believe Kelly Criterion was originally for use with fixed odds, like blackjack, and that, in my opinion, is where the difficulty comes in with horse racing. Everyone I know who claimed to use Kelly Criterion was using a fraction (1/2 - 3/4) of Kelly bet size.

[Dave Schwartz]

Nobody on God's green earth uses the Kelly Criterion to size their bets. Nobody.

Actually, that is not a true statement.

I know many people who SIZE their bets using Kelly. However, they do not bet that percentage of BR. They bet the full percentage of a FIXED BR size.

This way the bets are of the right RELATIVE size to each other in relationship to the risk.

[andicap]

Actually, that is not a true statement.

I know many people who SIZE their bets using Kelly. However, they do not bet that percentage of BR. They bet the full percentage of a FIXED BR size.

This way the bets are of the right RELATIVE size to each other in relationship to the risk.

Can u give an example? I recall in one your seminars you said had changed your mind about Kelly.

[Dave Schwartz]

Sure. Let's say that you have a $50,000 bankroll. (I know, but bear with me.)

You decide to size each wager as if the BR is $5,000.

Each wager is proportionately sized according to solid risk-reward principles.

Your larger bets will be on horses that deserve to have more on them.

As you win or lose during the day, the BR is always static at $5k.

[lansdale]

Nobody on God's green earth uses the Kelly Criterion to size their bets. Nobody.

I've written this a few times on this site, but nobody seems to be listening. Some of the most successful handicappers of all time used and continue to use Kelly. It is the optimal way to bet, proven beyond all doubt. However, extremely few horseplayers have the training and education to use it correctly. That's not the same as 'nobody'.

[Dave Schwartz]

I've written this a few times on this site, but nobody seems to be listening. Some of the most successful handicappers of all time used and continue to use Kelly. It is the optimal way to bet, proven beyond all doubt. However, extremely few horseplayers have the training and education to use it correctly. That's not the same as 'nobody'.


Lansdale,

Actually, they use it as I defined above.

[lansdale]

Sure. Let's say that you have a $50,000 bankroll. (I know, but bear with me.)

You decide to size each wager as if the BR is $5,000.

Each wager is proportionately sized according to solid risk-reward principles.

Your larger bets will be on horses that deserve to have more on them.

As you win or lose during the day, the BR is always static at $5k.

Hi Dave,

Not quite sure what you mean here. At first, with the 5k, I thought you were describing someone betting 1/10 Kelly. But I'm not sure how that jibes with, 'Each wager is proportionately sized according to sold risk-reward principles. Your larger bets will be on horses that deserve to have more on them.' This sounds like simply betting more on lower odds horses. This isn't Kelly, as I'm sure you know.

Cheers,

lansdale

[Dave Schwartz]

Let me simply ask you a question.

You have a player with a $5m in his account and another $50m in assets. Do you really think that when he loses two bets in a row, his bets shrink?

I assure you that the players you describe MAX their bets RELATIVE TO THE POOL, NOT THEIR ADW account balance.


Not quite sure what you mean here. At first, with the 5k, I thought you were describing someone betting 1/10 Kelly. But I'm not sure how that jibes with, 'Each wager is proportionately sized according to sold risk-reward principles. Your larger bets will be on horses that deserve to have more on them.' This sounds like simply betting more on lower odds horses. This isn't Kelly, as I'm sure you know.

Optimum bet is advantage / odds * bankroll.

This does not AUTOMATICALLY bet more on low-priced horses, although that is often the result.

[lansdale]

Let me simply ask you a question.

You have a player with a $5m in his account and another $50m in assets. Do you really think that when he loses two bets in a row, his bets shrink?

I assure you that the players you describe MAX their bets RELATIVE TO THE POOL, NOT THEIR ADW account balance.




Optimum bet is advantage / odds * bankroll.

This does not AUTOMATICALLY bet more on low-priced horses, although that is often the result.

Dave,

No disagreement about the constraint of pool-size limits for big bettors in big pools or average or small bettors in small pools - Benter covered this also. So you really are talking about betting 1/10 Kelly, and you agree that the key to the bet is the edge (what Benter also calls the 'advantage', and what Delta Lover calls the 'magnitude') and that the odds are irrelevant. There seems to much disagreement in this thread about something that seems simple and is usually undisputed. Thanks for clearing this up.

Cheers,

lansdale

[Dave Schwartz]

No. Because in a 1/10 Kelly (or any fractional Kelly) the BR changes with wins or losses.

I am saying that you size your wagers based upon a FIXED BR size.

This is actually a great way to play because you size your wagers RELATIVE to risk/reward (which is what Kelly is actually about, of course).

My experience is that most players misunderstand the proper USE of Kelly. They think it is to maximize growth when it is really to simply apply good risk/reward principles.

Listen, all hit rates (and odds these days) are really nothing but a good guess anyway. There is just no point in trying to size your bets relative to TOTAL bankroll.

In this new thread which DL started, I think that technically both guys are correct.

Paraphrasing...
The one guy says, "I don't think anyone is using it in the real world" - I agree. Nobody is wagering percentage of true bankroll.

The other guys says, "But the big boys do use Kelly." - I agree, but not in the strictest sense.

And another thing... what is the definition of "bankroll?" Is it all the money in your wallet, your bank account; your net worth? Of course not. It is whatever you have currently committed to risk.

The whales are (effectively) "flat betting Kelly."


Regards,
Dave Schwartz


Regards,
Dave Schwartz

[lansdale]

No. Because in a 1/10 Kelly (or any fractional Kelly) the BR changes with wins or losses.



I am saying that you size your wagers based upon a FIXED BR size.



This is actually a great way to play because you size your wagers RELATIVE to risk/reward (which is what Kelly is actually about, of course).



My experience is that most players misunderstand the proper USE of Kelly. They think it is to maximize growth when it is really to simply apply good risk/reward principles.



Listen, all hit rates (and odds these days) are really nothing but a good guess anyway. There is just no point in trying to size your bets relative to TOTAL bankroll.



In this new thread which DL started, I think that technically both guys are correct.



Paraphrasing...

The one guy says, "I don't think anyone is using it in the real world" - I agree. Nobody is wagering percentage of true bankroll.



The other guys says, "But the big boys do use Kelly." - I agree, but not in the strictest sense.



And another thing... what is the definition of "bankroll?" Is it all the money in your wallet, your bank account; your net worth? Of course not. It is whatever you have currently committed to risk.



The whales are (effectively) "flat betting Kelly."



Regards,
Dave Schwartz

Dave,

First, can we agree that this is largely an academic issue - few players have the knowledge to use Kelly effectively#

Second, based on the opinion of many here, even for those that know how to use it, the current environment of late-odds fluctuation have rendered Kelly useless#

That said, Kelly *is* the optimal technique for maximizing bankroll growth# There's no argument about this among the math people# Of course it incorporates 'risk vs# reward' principles, but so do a thousand other horseplayer money-management schemes many of which have proven worthless#

If at this point, even those who would have been capable of building a model and using Kelly in the more stable odds environment of the past would be unable to use it in today, and even the best models can provide no more than a 'good guess', than one can't call the bets made on such a weak foundation Kelly bets# And if a 'good guess' is all they are capable of providing, than it does make sense to be flat-betting - but why then insist on using the word 'Kelly' in conjunction with this kind of play?

Re the definition of 'bankroll' - this comes up often among BJ players - and most consider it 'the total amount one is willing to lose#'

Cheers,

lansdale

[lansdale]

I tried to post a couple of times here - subject lines registered, but not body copy. Another test.

[MightBeSosa]

folks have been talking about kelly and horseracing for decades.

imo, it has very little to do with profitability in the real world. imo , its invoked by those that think some magic formula will cure their lousy handicapping and betting skills.

kelly is fine for those already at a nice r.o.i and who wish to maximize gain, not meant to be used by losers.

[lansdale]

Seems like thngs are working again, so, I'll try to re-post.

Brief response to Dave's last post:

Agree with what you say, in general, but something as vague as 'good guesses' can't work with Kelly. Thus, it makes sense that the whales, in dealing with wildly fluctuating odds are therefore flat-betting, if all they have to work with are 'good guesses', but this is not Kelly, which depends on accuracy. In this context, I think Kelly can be more valuable for testing purposes. If you don't have an edge, Kelly will very quickly reveal it. For every 'real' bet you make, make Kelly (or fractional Kelly) paper bets, and see how you would have done. It would seem that it could also be used to test weaknesses.

Cheers,

lansdale

[PaceAdvantage]

I tried to post a couple of times here - subject lines registered, but not body copy. Another test.It was working...you were just messing up the QUOTE tags on Dave's quote...I fixed it for you in your first reply.

[lansdale]

It was working...you were just messing up the QUOTE tags on Dave's quote...I fixed it for you in your first reply.

Appreciate the headsup and correction.

lansdale

[Dave Schwartz]

Lansdale,

You misunderstand.

They are using Kelly as I described. What I meant by "flat betting Kelly" is that it is not in relation to a FLUCTUATING bankroll, but a STATIC one.

IOW, they are using Kelly as a means to balance risk and reward on their wagers.

[lansdale]

Lansdale,

You misunderstand.

They are using Kelly as I described. What I meant by "flat betting Kelly" is that it is not in relation to a FLUCTUATING bankroll, but a STATIC one.

IOW, they are using Kelly as a means to balance risk and reward on their wagers.

Dave,

Not trying to be difficult here but I truly do not understand what you are trying to say. The betting pattern you describe may be working well, but it has nothing to do with Kelly as defined. Just to address one point, you know that there is no such thing as a 'static' bankroll from the perspective of Kelly. Most blackjack players, the vast majority of whom use Kelly, consider the loss of 50% of bankroll to be the time to cut the bet-size in half. If you look at Magister Ludi's response to Robert Fischer in the 'Kelly' thread, about breaking his total bank into separate bankrolls, he makes somewhat the same point. I in no way believe that Kelly is the only to to go, and I think that most agree on how difficult it is to use in practice, but how does what you describe relate to Kelly?

Cheers,

lansdale

[Dave Schwartz]

I guess then, we will just have to disagree.

[Robert Fischer]

Lansdale ,
I think what Dave is talking about is using Fractional Kelly, but only updating the bankroll amount for each day(or a stop-loss point), not necessarily each play.

correct me if i'm wrong.

[Dave Schwartz]

Close, but NEVER change the BR amount and use FULL Kelly.

[raybo]

Close, but NEVER change the BR amount and use FULL Kelly.

Dave,

Are you saying that, because the operating bankroll ($5000) never changes, then your bet size never changes, unless your "advantage" changes? I'm confused also.

[Dave Schwartz]

The bet size changes as the probabilities and odds on the horses are different.

[lansdale]

I guess then, we will just have to disagree.

Hi Dave,

You say we disagree, but unlike much that is subjective in the realm of handicapping, the mathematics of gambling are universal and objective, not a matter of opinion. The Kelly Criterion and its use are simple and clearly defined, and you still haven't made clear how the betting technique you describe, betting full Kelly on an ever-replenished bank, relates to it. 'Good guesses' can't be the basis of Kelly bets.

Cheers,

lansdale

[Dave Schwartz]

I will try one more time.



Race 1
====
Bankroll = $5,000

Let's say you have a race with two overlaid horses:

A) 4/1 25% $Net = $2.50 opt%=6.25% Bet=$312
B) 8/1 13% $Net = $2.34 opt%=2.13% Bet=$106


Whether the race is a winner or a loser, the BR remains the same after the race is over.

Race 2
====
Bankroll = $5,000 (still and always)

Let's say you have a race with one overlaid horse:

A) 1/2 82% $Net = $2.46 opt%=28.75% Bet=$1,437


Race 3
====
Bankroll = $5,000 (still and always)

Let's say you have a race with one overlaid horse:

A) 2/1 35% $Net = $2.10 opt%=5.00% Bet=$250



So, in each race the bets have been SIZED to represent proper risk reward based upon (what you BELIEVE to be) the correct hit rate(s) and odds.

At the end of the day (or week, or month) you will have wagered more money on the races/horses that CALLED FOR larger bets and less on the ones that called for smaller bets.

[Magister Ludi]

When pools are relatively small and/or bankrolls are relatively large, there is an optimal amount to bet on an overlay in its respective parimutuel pool. Example:

assume that bankroll = $10,000,000

#__est. odds__est. prob___pool_____act. odds__optimal bet
1___4_________.20________$25,725____.175______$187 8.66
2___4_________.20________$22,050____.15_______$332 9.33
3___9_________.10________$ 7,350____.05_______$2845.77
4___1.5_______.40________$73,500____.50_______$0
5___9_________.10________$18,375____.125______$0
TOTALS_____1.00______$147,000__1.00

Betting full Kelly would cause the overlays to become underlays. The above optimal bets extract the maximum expected value from the pool.

[raybo]

The bet size changes as the probabilities and odds on the horses are different.

Ok, that makes more sense. But, still, your static $5000 operating bankroll is only static because you continue to add to it or subtract from it, from your fluctuating total bankroll. Whatever, if it accomplishes your goal, then, so be it.

[Dave Schwartz]

"Static" means "doesn't change." There is no adding or subtracting.

[raybo]

"Static" means "doesn't change." There is no adding or subtracting.

You said it was a static $5000. How does it remain static when it becomes less than $5000? I'm confused now. What's the purpose of the total bankroll then? Or, do you just play until the $5000 disappears and then start over with a new $5000? :confused:

[Dave Schwartz]

The purpose of the bankroll is to determine how much a "good bet" is worth.

For example, you may find that in a high-risk, low-return race a typical bet might be (say) $50-75 while a low-risk, high-return race might result in a bet of $400-$500.

Put it in a spreadsheet and try some examples. You will see what I mean.

Here... try these...

A) 2/1 @ 35% $Net=$2.10 Opt%= 5.0% Bet=$250
B) 2/1 @ 40% $Net=$2.40 Opt%=10.0% Bet=$500.
C) 2/1 @ 45% $Net=$2.70 Opt%= 17.5% Bet=$875

See how it scales has the hit rate goes up or down?

D) 5/1 @ 18% $Net=$2.16 Opt%= 3.6% Bet=$180
E) 5/1 @ 20% $Net=$2.40 Opt%= 4.0% Bet=$200
F) 5/1 @ 22% $Net=$2.64 Opt%= 6.4% Bet=$320

G) 4/5 @ 58% $Net=$2.09 Opt%= 5.6% Bet=$280
H) 4/5 @ 60% $Net=$2.16 Opt%= 10.0% Bet=$500
I) 4/5 @ 80% $Net=$2.88 Opt%= 35.2% Bet=$1,760

So, these numbers tell us that we'd rather have a 2/1 horse with a 45% chance of winning ($875) than a 5/1 @ 22% ($320).

[raybo]

The purpose of the bankroll is to determine how much a "good bet" is worth.

For example, you may find that in a high-risk, low-return race a typical bet might be (say) $50-75 while a low-risk, high-return race might result in a bet of $400-$500.

Put it in a spreadsheet and try some examples. You will see what I mean.

Here... try these...

A) 2/1 @ 35% $Net=$2.10 Opt%= 5.0% Bet=$250
B) 2/1 @ 40% $Net=$2.40 Opt%=10.0% Bet=$500.
C) 2/1 @ 45% $Net=$2.70 Opt%= 17.5% Bet=$875

See how it scales has the hit rate goes up or down?

D) 5/1 @ 18% $Net=$2.16 Opt%= 3.6% Bet=$180
E) 5/1 @ 20% $Net=$2.40 Opt%= 4.0% Bet=$200
F) 5/1 @ 22% $Net=$2.64 Opt%= 6.4% Bet=$320

G) 4/5 @ 58% $Net=$2.09 Opt%= 5.6% Bet=$280
H) 4/5 @ 60% $Net=$2.16 Opt%= 10.0% Bet=$500
I) 4/5 @ 80% $Net=$2.88 Opt%= 35.2% Bet=$1,760

So, these numbers tell us that we'd rather have a 2/1 horse with a 45% chance of winning ($875) than a 5/1 @ 22% ($320).

Ahhhh! Gotcha now. :ThmbUp:

[formula_2002]

Dutching is stressful l tool for dulling the pain over a long period of time compared to dying suddenly.
In a sample as small as 194 races, a proportional odds bet on the fav returned a .67 roi
A flat be on the favorite returned .66 roi.
The second choice proportional odds bet returned a .94 roi, a flat bet roi of .96
The third choice proportional odds bet returned a 1.03 roi, a flat bet roi of1.04

If you dutched the 2nd and 3rd choices the roi would be .985, a flat bet roi of 1.00

The point is this, in a sample of only 194 races, it’s shown that a Dutch, proportional or flat bet returns appox the same roi’s for the 1st, 2nd and 3rd choices.

In the same sample, a bet on all horses returned a .78 dutch bet roi and a flat bet roi of .81

The average book percentage in the above was 1.22