Logistic Model Question

[Sapio]

I recently re-read Benter's paper on a logistic model for horse racing. In his paper he says,"delta R2 of .0178 indicates that significant profits could be made with that model", where delta R2 is the difference between the public and the fundamental model.

Question 1: Is it plausible that a small difference actually leads to significant profits?

Question 2: I have also read in several threads in this forum about "reshaping" the public odds-line to achieve a better odds-line. Is this plausible?

Thomas Sapio

[GameTheory]

Originally Posted by Sapio

I recently re-read Benter's paper on a logistic model for horse racing. In his paper he says,"delta R2 of .0178 indicates that significant profits could be made with that model", where delta R2 is the difference between the public and the fundamental model.

Question 1: Is it plausible that a small difference actually leads to significant profits?

Question 2: I have also read in several threads in this forum about "reshaping" the public odds-line to achieve a better odds-line. Is this plausible?

Thomas Sapio

Considering the amount of money Benter made, it is more than plausible. He was working in Hong Kong, of course...

[DeltaLover]

Inspired by the notorious Bolton – Chapman parer, I have implemented several variants of Multinomial Logit Models. At this moment I do not recall the exact values of R2 that I was able to achieve, but what matters is that I have found this approach to be incapable of beating the game. Although true that a (relatively) small difference in the R2 might lead to significant increase of the final PNL, still it seems that it is impossible to actually beat the takeout and start showing a profit.

As the applied model improves by using the most significant factors it tends to mimic the behavior of the crowd and deviations from the public becomes more and more sparse.

This is not to mean that such a model is useless, as been able to predict crowd's opinion is not an easy task and to some extend this is what handicapping is all about. What makes the difference for betting purposes is not the handicapping itself but the meta layer which can be seen as handicapping the handicappers.

[acorn54]

i'll add my two cents.the black magic software, marketed by mike pizzolla has as part of its functionality the programs projection of what the public odds of each horse in a race will be at post time. i was impressed that a strictly- by- the numbers approach computer program, could be as accurate as blam was at predicting the crowds behavior.

[Sapio]

Originally Posted by DeltaLover

Inspired by the notorious Bolton – Chapman parer, I have implemented several variants of Multinomial Logit Models. At this moment I do not recall the exact values of R2 that I was able to achieve, but what matters is that I have found this approach to be incapable of beating the game. Although true that a (relatively) small difference in the R2 might lead to significant increase of the final PNL, still it seems that it is impossible to actually beat the takeout and start showing a profit.

As the applied model improves by using the most significant factors it tends to mimic the behavior of the crowd and deviations from the public becomes more and more sparse.

This is not to mean that such a model is useless, as been able to predict crowd's opinion is not an easy task and to some extend this is what handicapping is all about. What makes the difference for betting purposes is not the handicapping itself but the meta layer which can be seen as handicapping the handicappers.

Hi DL

"What makes the difference for betting purposes is not the handicapping itself but the meta layer which can be seen as handicapping the handicappers."

Is the above statement equivalent to handicapping the odds?

Thomas Sapio

[arkansasman]

Originally Posted by Sapio

I recently re-read Benter's paper on a logistic model for horse racing. In his paper he says,"delta R2 of .0178 indicates that significant profits could be made with that model", where delta R2 is the difference between the public and the fundamental model.

Question 1: Is it plausible that a small difference actually leads to significant profits?

No

[Robert Goren]

To put it in simple terms, a delta R2 of .0178 is huge where y is either 1 or 0 and there are several 0 for each 1. If you can get that and it is real and not just a statistical fluke, you will get very rich.

[Sapio]

Originally Posted by acorn54

i'll add my two cents.the black magic software, marketed by mike pizzolla has as part of its functionality the programs projection of what the public odds of each horse in a race will be at post time. i was impressed that a strictly- by- the numbers approach computer program, could be as accurate as blam was at predicting the crowds behavior.

Hi acorn

There appears to be several software packages that claim to model the crowds behaviour, but they fail to do better than the crowd. Why is that?

Thomas Sapio

[Sapio]

Originally Posted by arkansasman

No

Hi arkansasman

Is that a no based on experience or a gut feeling type no?

Thomas Sapio

[arkansasman]

Hi Thomas,

It is from experience.

John

[GameTheory]

Your question #2 about reshaping the oddslines is the important one. If you don't do that, no you won't be making any money with this approach...

[acorn54]

Originally Posted by Sapio

Hi acorn

There appears to be several software packages that claim to model the crowds behaviour, but they fail to do better than the crowd. Why is that?

Thomas Sapio

well, in the black magic software the point of projecting the crowds behavior in terms of their probability of each horse's chances in the race, is so you can use that as a guide as to what the blam odds line is for each horse and bet those horses that blam see's as having a better chance than the projected post time odds.
i hope that makes sense. whether or not this idea in the black magic software works, i can not honestly say, as i only owned black magic for a month before selling it because i found the betting set-up i was using before purchasing it was satisfactory for me.

[Sapio]

Originally Posted by arkansasman

Hi Thomas,

It is from experience.

John

Hi John,

As GameTheory correctly pointed out Benter was operating in HK. Possibly, a delta of .0178 was significant. In your experience with NA tracks, what level of delta R2 would you consider to be significant?

Thomas Sapio

[DeltaLover]

Originally Posted by Sapio

Hi DL

"What makes the difference for betting purposes is not the handicapping itself but the meta layer which can be seen as handicapping the handicappers."

Is the above statement equivalent to handicapping the odds?

Thomas Sapio

Not necessary!

What I am trying to say here, is that instead of basing our selection process on comparing the odds – line as offered by the public to what we consider to be our accurate prediction, we start from classifying a race (viewed as a whole) as an event with high probability for the public to commit a large mistake betting on it.

Shifting our focus from the individual horse to the whole race and having an opinion of how possible it is for the crowd to misjudge it, means that we start with an priory assumption that helps us answer the most fundamental question a horse bettor is facing: TO BET OR NOT THE SPECIFIC RACE

Going further we can become more specific about the source of the betting inefficiency concentrating in specific areas of the board. Meaning our system can suggest that there is a high probability the inefficiency to exist on over betting the top choices, something that will make us look for value among the middle of the board or vice versa...

I any way the important thing here is, that we do not try to estimate each horse individually but to have a more generic opinion.

[arkansasman]

Thomas,

At a delta R2 > .009, you should see a larger number of advantage bets.

John