Handicapping Using Similarities And Dissimilarities

[DeltaLover]

Originally Posted by baconswitchfarm

I don't know which edition I have. It is a green hardcover with gold lettering and looks old. It is a really slow tedious read. I would say it helped me a small bit just looking at things differently. It is not a light read for when you are falling asleep in bed. It took some focus for me to commit to finishing it.

Please let's keep the thread on target

[Robert Goren]

Start with first time starters which basically have the PPs (none). The well bet ones win a lot more than the 20/1 ones. The reason is stable money and stable gossip and public jumping on the early betting.

[DeltaLover]

Originally Posted by Robert Goren

Start with first time starters which basically have the PPs (none). The well bet ones win a lot more than the 20/1 ones. The reason is stable money and stable gossip and public jumping on the early betting.

FTSs have their own handicapping angles that can be used for clustering..

This is not the main point though...

The challenge is how can we qualify a cluster... In other words, taking your example of FTO starters, starting from the assumption that all starters who are from a sire with a stud fee over 25K also showing a bullet work consist a cluster, how can I prove it? How can I be sure that every group of such starters within a race are winning with the same frequency?

I am under the impression that if we can solve this problem we will be closer to a winning strategy than we were before...

[eurocapper]

I think horses are individual to a large degree and get even more so during their career with trainer switches, injuries etc. Horses from the same sire running in a maiden with similar appearance and similar prior workouts may be the best cluster one can get. I believe there may be value there, since most pedigree handicapping for example focus only the sire not the dam.

[DeltaLover]

If this is the case, then we need to know if this clustering is good or not

This is exactly what I am looking for..

Having this cluster how can we conclude that indeed all its starters win with the same frequency thus consisting a 'good' cluster?

[eurocapper]

Originally Posted by DeltaLover

If this is the case, then we need to know if this clustering is good or not

This is exactly what I am looking for..

Having this cluster how can we conclude that indeed all its starters win with the same frequency thus consisting a 'good' cluster?

No it would depend on the competition obviously (and if the trainer just wants to workout at the maiden etc.). For win percentage specifically I don't know if there is a better factor than a "strategic" trainer who is trying to keep up his win percentage and is selective about races and scratches etc., but not much value there. Of course sometimes their horses lose too.

[DeltaLover]

I do not understand what you mean by depending on the competion..

I view it like this:

A Race can be broken down to three clusters:

A -> constists from 5 horses
B -> constists from 3 horses
C -> constists from 3 horses

For my cluster to be worth of value, all starters in each individual one should win with the same frequency. Otherwise we are missing something.

How can we be sure that each starter in each cluster A,B,C is winning with the same frequency?

[eurocapper]

Originally Posted by DeltaLover

I do not understand what you mean by depending on the competion..

I view it like this:

A Race can be broken down to three clusters:

A -> constists from 5 horses
B -> constists from 3 horses
C -> constists from 3 horses

For my cluster to be worth of value, all starters in each individual one should win with the same frequency. Otherwise we are missing something.

How can we be sure that each starter in each cluster A,B,C is winning with the same frequency?

Ok I missed that these clusters are within a race and not horse clusters in general. In my opinion for that approach some kind of win-% at that class level might be best for this kind of approach. If there is some doubt about the win-%, why not use it directly.

[formula_2002]

Originally Posted by Dave Schwartz

I have both of his books in my collection.

Neither are worth the effort to read them IMHO.

Worth the read if only for his sound analytical approach
It has saved me countless dollars..

[DeltaLover]

Originally Posted by formula_2002

Worth the read if only for his sound analytical approach
It has saved me countless dollars..

Formula, you really think your comment is on topic?

[formula_2002]

Originally Posted by DeltaLover

Formula, you really think your comment is on topic?

Yes, only if we are writing about "Handicapping Using Similarities And Dissimilarities"

Fabricand wrote the book on the subject... including the analysis..

[baconswitchfarm]

I also mucked up the thread by talking about the book written about the exact topic of the thread. In my defense , there were four references to it on the previous page that didn't get chastised. My apologies, please continue.

[shouldacoulda]

I don't know if this helps but I try to break it down to contenders and non contenders first. From there I look at the contenders for a standout, or angle with an overlay. If there is no angle, standout or overlay I pass the race. There are too many races at too many tracks to split hairs on a given race. To me it's just not worth the time otherwise.

[DeltaLover]

Originally Posted by shouldacoulda

I don't know if this helps but I try to break it down to contenders and non contenders first.

Grouping starters to contenders and not, is an approach all of us have followed at some time and a lot of us still start our handicapping from there...

Besides this, I think this heuristic is hiding several loopholes that need to be addressed and resolved.

The question is what do we really mean by a contender? Or equally what we mean by an non contender?

The chaotic nature of the game implies that every starter in a race has some chance to win it. It is impossible to create a selection process to select no winners out of a significant sample. The lowest I was able to go was in the range of 3% for races consisting from 7 to 10 starters. The precise number is not important here, what matters is the fact that it is wrong to assign to any starter (even more to a group of them) their winning chance to be zero.

Based in this I am very skeptic about such a grouping and see it more like an ad hoc approximation that might have some value under ideal circumstances but overall cannot serve its purpose.

At this point I want to emphasise the fact that our negative opinion about a starter is more valuable the more this horse is preferred by the public. Having a negative opinion about a 100-1 shot does not help us much... We would rather have this kind of an opinion of a 1-5 shot....

Symmetrically we can make the reverse statement for positive opinions.

So, we are back to our standard challenge: How to beat public's opinion...

I am opening a parenthesis here.

I have seen several times written here in PA and also I have heard it many more times at the track that in this race the public had it wrong, since the 8-5 favorite lost to this 10-1 long shot.

This statement is a first class fallacy encapsulating a very weak understating of the game.

To illustrate this concept, assume the following:

In an fictitious world, we are presented of a sequence of 10 races each one consisting of 10 starters who only can bet to win. For simplicity let's assume that there is no take out at all. It happens that all of them are won by 9-1 shots. W

hat can we say about how the public had them? Was it's opinion correct or wrong?

We simply do not have enough data to answer this question.

If instead of 9-1 we had all the winners at 20-1 then yes we could immediately conclude that the crowd indeed did not do a good job.

Let's now assume that all the winners came back at even money. What do you think about this crowd's opinions?

Somebody might be tempted to jump to an easy conclusion that this crowd is much tougher and accurate in its line making...

Is it like this though?

No it is not..

This case where all winners come back as even money winners represent a much softer than the first one where all are 9-1

Can you see why?

Just because the former case implies that the crowd is underestimating the chances of the even money favorite which always wins.

Note that it is impossible to have two even money shots in a race of 10. Always there will only be one such a horse. So based in the outcome of the 10 races this one horse was always undervalued. This crowd is easy to take advantage of. Add a single more bettor who always bets the horse 'selected' by the pool as even money,. He will end up taking all their money. Please note that this is not possible in the first case.

I think this 'toy' example proves that the final odds of the winner do not dictate how good the crowd bets or not and if it had it correctly or not.

Here let me close the parenthesis and go back to the original topic of diving a race to contenders and non contenders.

Instead of looking for them it seems reasonable to look for situations where the crowd behaves more that the case of even money shots rather than 9-1 in our example. Since the public's logic is based in past performances it makes sense to try to detect patterns that tend to mislead it to wrong conclusions.

Here, let's assume that we have a race-level metric expressing the similarity of two races. At this point I am not concerned about how this metric is created. I only care about what is measuring.

If this metric is able to come up with some groups of races resembling the even money races in our example then our task as bettors is simplified. We already have find our target group. We just ignore all other races concentrating in these that we know before hand that the public has them wrong.

This exactly is the value that has to be offered by this similarity metric.

The next challenge we are facing is the how.

What are the attributes that we need to estimate to arrive to this metric?

How do we process them and decide what is important and what is not?

What should be the shape of the data we should deal with, in order to simplify the whole task?

These are some of the challenges we are facing and their resolution is the topic of this thread....

What do you think?

[thaskalos]

Originally Posted by DeltaLover

Grouping starters to contenders and not, is an approach all of us have followed at some time and a lot of us still start our handicapping from there...

Besides this, I think this heuristic is hiding several loopholes that need to be addressed and resolved.

The question is what do we really mean by a contender? Or equally what we mean by an non contender?

The chaotic nature of the game implies that every starter in a race has some chance to win it. It is impossible to create a selection process to select no winners out of a significant sample. The lowest I was able to go was in the range of 3% for races consisting from 7 to 10 starters. The precise number is not important here, what matters is the fact that it is wrong to assign to any starter (even more to a group of them) their winning chance to be zero.

Based in this I am very skeptic about such a grouping and see it more like an ad hoc approximation that might have some value under ideal circumstances but overall cannot serve its purpose.

At this point I want to emphasise the fact that our negative opinion about a starter is more valuable the more this horse is preferred by the public. Having a negative opinion about a 100-1 shot does not help us much... We would rather have this kind of an opinion of a 1-5 shot....

Symmetrically we can make the reverse statement for positive opinions.

So, we are back to our standard challenge: How to beat public's opinion...

I am opening a parenthesis here.

I have seen several times written here in PA and also I have heard it many more times at the track that in this race the public had it wrong, since the 8-5 favorite lost to this 10-1 long shot.

This statement is a first class fallacy encapsulating a very weak understating of the game.

To illustrate this concept, assume the following:

In an fictitious world, we are presented of a sequence of 10 races each one consisting of 10 starters who only can bet to win. For simplicity let's assume that there is no take out at all. It happens that all of them are won by 9-1 shots. W

hat can we say about how the public had them? Was it's opinion correct or wrong?

We simply do not have enough data to answer this question.

If instead of 9-1 we had all the winners at 20-1 then yes we could immediately conclude that the crowd indeed did not do a good job.

Let's now assume that all the winners came back at even money. What do you think about this crowd's opinions?

Somebody might be tempted to jump to an easy conclusion that this crowd is much tougher and accurate in its line making...

Is it like this though?

No it is not..

This case where all winners come back as even money winners represent a much softer than the first one where all are 9-1

Can you see why?

Just because the former case implies that the crowd is underestimating the chances of the even money favorite which always wins.

Note that it is impossible to have two even money shots in a race of 10. Always there will only be one such a horse. So based in the outcome of the 10 races this one horse was always undervalued. This crowd is easy to take advantage of. Add a single more bettor who always bets the horse 'selected' by the pool as even money,. He will end up taking all their money. Please note that this is not possible in the first case.

I think this 'toy' example proves that the final odds of the winner do not dictate how good the crowd bets or not and if it had it correctly or not.

Here let me close the parenthesis and go back to the original topic of diving a race to contenders and non contenders.

Instead of looking for them it seems reasonable to look for situations where the crowd behaves more that the case of even money shots rather than 9-1 in our example. Since the public's logic is based in past performances it makes sense to try to detect patterns that tend to mislead it to wrong conclusions.

Here, let's assume that we have a race-level metric expressing the similarity of two races. At this point I am not concerned about how this metric is created. I only care about what is measuring.

If this metric is able to come up with some groups of races resembling the even money races in our example then our task as bettors is simplified. We already have find our target group. We just ignore all other races concentrating in these that we know before hand that the public has them wrong.

This exactly is the value that has to be offered by this similarity metric.

The next challenge we are facing is the how.

What are the attributes that we need to estimate to arrive to this metric?

How do we process them and decide what is important and what is not?

What should be the shape of the data we should deal with, in order to simplify the whole task?

These are some of the challenges we are facing and their resolution is the topic of this thread....

What do you think?

Can you simplify this somewhat...for the benefit of someone to whom English is a second language?