The only one absolute truth toward profitability

[Overlay]

Originally Posted by BELMONT 6-6-09

I remember reading this one statement when I first started taking the races seriously: HERE IS THE ANSWER-DON'T REGARD IT LIGHTLY-IT IS THE WHOLE SECRET: these professionals win because they know the inside principle of beating the races, the same principle that must be used to beat any speculative game or business from which a legal take, house percentage or brokerage fee is extracted,the principle is "COPPER THE PUBLIC'S IDEAS AND PLAY AT ALL TIMES! this is not abstract theory-it is practical percentage...copper meaning to bet against.

This statement was from The secrets of professional turf betting by Robert L. Bacon

I agree that you can't conform completely to the public's handicapping and be a long-term winner, but I would also take issue with totally disregarding the public's opinion and reasoning simply for the sake of being contrarian. There are still established performance patterns and percentages associated with the game, and it becomes much more difficult to achieve a positive return if you consistently bet against those realities. I would modify the principle above to read, "Bet against the public's ideas and play when they're wrong, and when you can tell why they're wrong."

[CincyHorseplayer]

Originally Posted by Dave Schwartz

Your research shows different results than mine.

With one exception, I have never seen a single factor show profitability over (say) 3,000 wagers or so.

That one exception was that for several years the best of last 10 1st fraction rating was flat-bet profitable in claiming routes by 1%. That went away about 5 years ago.




Absolutely true. IMHO, there is no such thing as a horse with absolute, zero probability.




Exactly, and even more so. The profit of your wagering is dependant upon the non-contenders you select because they provide the part of the pool that is available for winning. In other words, in a race where the public has lined up the top 4 or 5 of YOUR contenders as the top public choices, especially if they are in the same OOP order, this will likely be a losing race in the long run for you, even if you manage to win today.

OOP = Order of preference.


"COPPER THE PUBLIC'S IDEAS AND PLAY AT ALL TIMES! this is not abstract theory-it is practical percentage...copper meaning to bet against.

IMHO, this will put your hit rate down in the very low range. If you can stand winning one race out of 8 or 10, then the game becomes relatively easy.

By the way, the term "Copper" actually comes from a game called Faro Bank (also called Buck-the-Tiger). And it does, in fact, mean to wager with the house.


Regards,
Dave Schwartz

In the context of the rest of my post Dave,that every race is unique,the single factor will always be changing.The race itself is in isolation.The handicapping single factor isn't,except as it solves this particular race.I'm not preaching any single factor as a universal truth.But that a single factor decides the race.Any one person,much less the public,won't land on it constantly.That's why I think it will always be profitable.That's why I'm optimistic about my chances.I've got Thaskalos to my left,Raybo to my right,and you and I comparing winning tickets thinking how we could have squeezed more blood from the race!At the end of the day were all winners while landing on completely different single factors.That's my vision of this game in a nutshell.I believe it.

[traynor]

Originally Posted by CincyHorseplayer

This is a single factor game.Every race is an isolated reality dominated at end by a single factor.This is why I think this game will always be profitable.There is a flood of good information out there.Sorting it out and finding the deciding factor on a regular basis will never happen collectively.

That is one of the most insightful observations I have read on this forum. I hope you pursue the matter further, expressing it in terms that enable others to readily understand its significance. Especially for those for whom your initial posting went completely over their heads.

[Dave Schwartz]

"Bet against the public's ideas and play when they're wrong, and when you can tell why they're wrong."

That is the key. Good post.

Most people, however, will not do the work. Instead, they go into a race saying, "Well, they made my top 2 horses the top public choices. Guess I will have to make another selection."

I KNOW this will be the key to my ultimate success. As was pointed out to me in my "Live Play" thread, I play too many of these races/horses myself.

Of course, in that thread I was testing my system (phase 1). Once I get it calibrated, I will begin to remove the plays that are costing me. But I need enough plays in the study to develop a set of rules.

[traynor]

Originally Posted by CincyHorseplayer

In the context of the rest of my post Dave,that every race is unique,the single factor will always be changing.The race itself is in isolation.The handicapping single factor isn't,except as it solves this particular race.I'm not preaching any single factor as a universal truth.But that a single factor decides the race.Any one person,much less the public,won't land on it constantly.That's why I think it will always be profitable.That's why I'm optimistic about my chances.I've got Thaskalos to my left,Raybo to my right,and you and I comparing winning tickets thinking how we could have squeezed more blood from the race!At the end of the day were all winners while landing on completely different single factors.That's my view of this game in a nutshell.I believe it.

The only clarification I might suggest is that people realize that the term "single factor" may be misleading. The "single factor" may be a compound of several factors, each of which might be (mislabeled and) understood as "single factors."

[Dave Schwartz]

In the context of the rest of my post Dave,that every race is unique,the single factor will always be changing.The race itself is in isolation.The handicapping single factor isn't,except as it solves this particular race.I'm not preaching any single factor as a universal truth.But that a single factor decides the race.Any one person,much less the public,won't land on it constantly.That's why I think it will always be profitable.That's why I'm optimistic about my chances.I've got Thaskalos to my left,Raybo to my right,and you and I comparing winning tickets thinking how we could have squeezed more blood from the race!At the end of the day were all winners while landing on completely different single factors.That's my vision of this game in a nutshell.I believe it.

Now THAT is something I can agree with wholeheartedly.

In fact, that is pretty close to my preferred approach.

I draw the best handful of factors from a limited set of factors.

[raybo]

Originally Posted by Dave Schwartz

Now THAT is something I can agree with wholeheartedly.

In fact, that is pretty close to my preferred approach.

I draw the best handful of factors from a limited set of factors.

By "set of factors" I assume you mean there may well be more "single" factors (a single "composite" factor), involved in the decision. Even though the factor(s) involved vary depending on the race and the field, etc., often a "set" of factors is involved versus only 1 individual factor.

[traynor]

Of course, it might be useful to understand that when regarding each race as a unique event that is not influenced by the races preceding it other than in relatively trivial ways, it eliminates the use of models constructed on other races as other than possible indicators, rather than having some (implied) cause-and-effect relationship on the outcome.

Specifically, the factor or combination of factors that prevail in the particular race in question are not derived from the results of the other races--it is the specific factor or factors that determine the winner of this race that matters, not the factor or factors that determined the winners of races yesterday or last week, because those races were also unique events.

It might take a bit to wrap one's head around, but the concept is useful. And because it goes so much against the grain of the way most bettors conceptualize races, it offers substantial opportunities for profit.

[Dave Schwartz]

Traynor,

I am not sure what you just said.

Could you explain it like I am a 5th grader, please?


Dave

[Dave Schwartz]

Raybo,

What I do is use an approach that we call The Dynamic PickMaster.

1. Open a race.

2. Apply a pre-designed Dynamic filter to the race.
This is a filter that makes filters according to a template. For example:

• Same Track, Surface Distance as this race
• Plus/minus 2 months of the year
• Similar track condition as this race (fst, good, off)
• Same age group (young or old)
• Maiden or non-maiden
• Similar pace pressure using (Giles' ES5 approach)

3. Run the filter against the database to capture races that fit.
If I do not have "enough" races then I pass the race.

4. Run the PickMaster against these races to create "objects."
The PickMaster:

• Creates 2 objects: 1 for early horses and 1 for late.
• uses a pre-selected list of factors (about 125 for me)
• can use different factors for early than late

The Pickmaster could be used to create other objects. For example, it can be used to determine which factors the public actually wagers upon, or which factors point to horses that will go to the front, or win (or run 2nd, etc.)

5. After these factors are selected (by the PkM) it creates an "Analyst" for the race which uses these objects.

The "Analyst" is a customized handicapping system, unique to this race, after scratches.

In my case, these objects are actually used to pick the contenders from the "NewPace" categories of Early and Late.

Some of our users have the PkM create a "Handicapping Object" as well as "contender Objects."


Note: All of this is accomplished with the click of a single button, and takes from 45 seconds to 4 minutes, depending upon speed of computer, number of races used (or found), etc.

[traynor]

Originally Posted by Dave Schwartz

Traynor,

I am not sure what you just said.

Could you explain it like I am a 5th grader, please?


Dave

The description of The Dynamic PickMaster seems to have it fairly well covered. The only thing I might add is that "unique event" also applies to the past races used in The Dynamic PickMaster. That is, the result of each of the races studied was a unique event, dependent on the existence of (or lack of) particular factors in each of the entries, not just in the winner.

Regression studies that only consider the attributes of the winner may be misleading, because the result of each race depends on the existence of (or lack of) particular factors in each of the entries, not just in the winner.

Example--closer wins this type of race, so look for a closer in this race (gross oversimplification for illustrative purposes). The attributes of the other entries in races used to determine the profile may be as important (and sometimes more important) than the attributes of the winner in each of the races used to create the profile. To get a clear picture, it is useful to study the losers as well as the winners when creating profiles. The winner profile may be perfectly valid when applied to similar races, but it may be deficient because it does not account for the existence or lack of other factors or combinations of factors in the losers.

It is those "confounding variables" that make modeling complex.

[bob60566]

Originally Posted by traynor

The description of The Dynamic PickMaster seems to have it fairly well covered. The only thing I might add is that "unique event" also applies to the past races used in The Dynamic PickMaster. That is, the result of each of the races studied was a unique event, dependent on the existence of (or lack of) particular factors in each of the entries, not just in the winner.

Regression studies that only consider the attributes of the winner may be misleading, because the result of each race depends on the existence of (or lack of) particular factors in each of the entries, not just in the winner.

Example--closer wins this type of race, so look for a closer in this race (gross oversimplification for illustrative purposes). The attributes of the other entries in races used to determine the profile may be as important (and sometimes more important) than the attributes of the winner in each of the races used to create the profile. To get a clear picture, it is useful to study the losers as well as the winners when creating profiles. The winner profile may be perfectly valid when applied to similar races, but it may be deficient because it does not account for the existence or lack of other factors or combinations of factors in the losers. It is those "confounding variables" that make modeling complex.

Would a factor be a losing horse who was being prepared for a future race by the trainer and he puts his jock on board for the losing effort and aquaint with the intention of getting back on board if the preparation is sucessful, that would make three factors but one in the next race.

[Dave Schwartz]

The description of The Dynamic PickMaster seems to have it fairly well covered. The only thing I might add is that "unique event" also applies to the past races used in The Dynamic PickMaster. That is, the result of each of the races studied was a unique event, dependent on the existence of (or lack of) particular factors in each of the entries, not just in the winner.

Traynor,

I absolutely agree.

I really think this is the Catch-22 of racing: All the races are unique, so how do you build a database of unique races?

I see it this way:
Let's say you build a system made up of the last 50,000 races in your database. Fast tracks, off tracks, dirt, turf, 2yr olds, older, maidens, graded stakes; everything.

Let's further say that your system, as developed, produces a -5% loss on the "best bet" in each race. (This is unimportant, so don't even give it another thought.)

If we were to create some kind of "similarity index," of the race shapes (not to be confused with pace), and plot that, we would probably find that our system did best in the middle part of the graph. (i.e. picture a Gaussian curve.)

Now, imagine as we begin to remove highly asimilar races from the sample, (i.e. turf, 2yr olds, off tracks, etc.) it would be logical that our system's results would improve.

I would summarize this by saying that the shape of the performance curve matches the degree of similarity between the races. Thus, as we continue to remove outliers from the sample (and rebuild the system), the performance would logically improve .

I believe the curve would begin to flatten as we continue to remove outliers.

This would be caused by the the greater similarity in the entire sample.

If we continue this process, our sample size dwindles, but the performance improves as we build a more specialized system.

Eventually, as the filtering template gets tighter and tighter, we reach the point where we do not have a "significant" sample. Secondarily, as we remove race types, if we want to play a wide spectrum of races we are forcing ourselves to create MANY MORE SYSTEMS!

I have had one user for several years who played over 1,200 systems. While HSH will handle that with relative ease, I am not sure that is the best way to do things.

This is why I chose to build "unique" systems. They are highly adaptive to change. BTW, on that topic, I ran through several hundred races a few weeks ago, searching for 100 "like" races. (I had to widen my filter a little to get these.) Then ran the same races again with a 50-race target. The 50-race method was actually superior to the 100-race method by a wide margin!

Now, understand that both methods go back several years, but, obviously, the 5-race method (on average) goes back only half as far (an average of about 3 years versus 6 years) to capture its sample.

I was very surprised at this result.

This tends to lead me towards the belief that data, like tomatoes in the grocery store, lose their usefulness with age.


Summary - Ultimately, every race is UNIQUE. No two races are EXACTLY alike. Logically, we must expand the filter. As we expand, we get more races, that are less alike, and have to go back further in time to get them.


There seems to be a dynamic relationship between:

• Tightness of filter
• Sample size
• Age of data

Hope this makes sense to somebody.


Regards,
Dave Schwartz

[traynor]

It makes a lot of sense, but because of the length, I won't quote it.

Short form--the more finely grained the base used to build the model, the more likely the model is to be replicated in future (similar) events. However, the more finely grained the base used to build the model, the less likely it becomes that future events will replicate events in the base, due to the existence or absence of the confounding variables mentioned previously.

I don't mean that to be simplistic or spurious. Conceptually, the relevance of past data as a predictor is limited to the range used in layering (how "finely grained" the match is). In the example you mentioned, accuracy increased in a 50 race base over a 100 race base. Without a pointless debate over sample sizes you might find it interesting to run a bootstrap algorithm (random sampling with replacement) on the 100 race (or whatever number) base to determine how "smooth" the model is.

Graphically, the bootstrap algorithm should a gentle wave slightly above and below a baseline that is fairly static. That would indicate a farily close correlation between the entire base and the (possible) future results. Conversely, a number of spikes and valleys would indicate anomalies within the segment studied that could tweak results and make the model less useful as a predictor.

[traynor]

Again, short form. When you determine your filtering criteria, it might be useful to avoid the Aristotelian model of most general to most specific as a means of categorization.

In racing, it seems perfectly reasonable (and in a simplistic sense, is perfectly reasonable) to assume that matches for a turf race should start by looking at other turf races, then layer by distance, and whatever else. I have done a lot of research that indicates such an approach may be misleading. As reasonable as it may seem superficially, the dynamics of a given race may be more significant predictors than (relatively unimportant) static details like surface, distance, etc.

By "dynamics" I am referring to "other factors" that may or may not include or exclude the primary criteria used to search for matches, and may or may not include or exclude specific confounding variables. It is quite possible that the "closest match" for a given race would be found in races with a similar mix of variables that do not necessarily include surface and distance.

That is one of the major difficulties in building models. If the basic assumption is that there is a progressive affect in a specific hierarchy of influences, that assumption will tend to exert a stronger influence on the dataset used for profiling than the details of the races extracted as "similar." It will also exclude races that may seem "dissimilar" but are in fact more representative.