Figuring coefficients in markets you control

[Dark Target]

When any of the larger syndicates start out they will either combine the log of the public's odds with their own, or combine the public odds with their independent variables to get their coefficients.

Initially when doing that process, the public's odds will be completely unaffected by their independent variables giving very good estimates.

Once the model is complete and they start betting into the pool, assuming the public odds are now significantly altered by their independent variables for most of the way through betting to the point of basically controlling the final odds, how would they obtain accurate coefficients going forward, given that now their independent variables presumably add nothing to the public estimates, because the public odds are now essentially their odds?

[Sapio]

When any of the larger syndicates start out they will either combine the log of the public's odds with their own, or combine the public odds with their independent variables to get their coefficients.

Initially when doing that process, the public's odds will be completely unaffected by their independent variables giving very good estimates.

Once the model is complete and they start betting into the pool, assuming the public odds are now significantly altered by their independent variables for most of the way through betting to the point of basically controlling the final odds, how would they obtain accurate coefficients going forward, given that now their independent variables presumably add nothing to the public estimates, because the public odds are now essentially their odds?

Hi Dark Target,

TM addressed this question in one of his threads. He was discussing the weight of the tote-odds dominating the fundamental model.

Thomas Sapio

[ReplayRandall]

Hi Dark Target,

TM addressed this question in one of his threads. He was discussing the weight of the tote-odds dominating the fundamental model.

Thomas Sapio
Hi Sapio,

Since TM has decided to no longer post here, could you possibly either find the thread you're referring and post it, and/or give us your own understanding of the key elements of the idea itself....Thank you.

ReplayRandall

[Sapio]

Hi Sapio,

Since TM has decided to no longer post here, could you possibly either find the thread you're referring and post it, and/or give us your own understanding of the key elements of the idea itself....Thank you.

ReplayRandall

Sorry, I don't have the exact thread (I have notes in my files), but I'm sure you can find it by searching TM and logistic regression.

Thomas Sapio

[Cratos]

Sorry, I don't have the exact thread (I have notes in my files), but I'm sure you can find it by searching TM and logistic regression.

Thomas Sapio
Hi Thomas,

Your absence along with TrifectaMike is conspicuously missed, but I think I understand why.

[Robert Fischer]

When any of the larger syndicates start out they will either combine the log of the public's odds with their own, or combine the public odds with their independent variables to get their coefficients.

Initially when doing that process, the public's odds will be completely unaffected by their independent variables giving very good estimates.

Once the model is complete and they start betting into the pool, assuming the public odds are now significantly altered by their independent variables for most of the way through betting to the point of basically controlling the final odds, how would they obtain accurate coefficients going forward, given that now their independent variables presumably add nothing to the public estimates, because the public odds are now essentially their odds?

I don't want to detract from the thread, and I am not criticizing it. Benter would probably like your question, and I hope you guys really dig in here, and possibly recreate some of the trifectamike stuff and have a hearty debate.

The way I see the game is a little different. I feel that I should know how the public is going to behave. I see and understand the system, so I know what is going to happen. Yet sometimes the public does not behave the way that I had expected. And that is generally because of a few things; randomness, an error of mine, or because of inside information.
Unless you are completely dominating and manipulating the market (say you wanted 4/5 on a horse that should be 3/5 so you make him open at 1/9), you are still able to predict and assess the market behavior prior to your wager.
And that is true whether you can do it in your head, or whether you understand the fundamental principles and have written equations to do it for you.
Obviously, if you are taking everything good out of a certain part of the market over a significant time frame, it will have an effect on the public's behavior, but that is a little different from what you are asking.

[DeltaLover]

When any of the larger syndicates start out they will either combine the log of the public's odds with their own, or combine the public odds with their independent variables to get their coefficients.


The problem with this approach is that there exist multiple syndicates that they bet against each other...

[Dark Target]

That is true, but what I mean is, when they start out they are comparing their coefficients to "clean" odds that are not impacted by them in anyway.

After each meeting as they refine their coefficients, if the syndicate is large enough to be betting odds down to the point they offer no value (according to them), then when you are comparing your coefficients to the public odds your are comparing them to odds that already take those coefficients into account.

[Dark Target]

Sorry, I don't have the exact thread (I have notes in my files), but I'm sure you can find it by searching TM and logistic regression.

Thomas Sapio

Thanks Sapio, I will see what I can find. Interested in your or others thoughts as to how this can be overcome also.

[sjk]

It seems to me you would want to understand how much your bet will affect the payoff. From there you could adjust your betting parameters to allow for the reduced payoff.

For example if you are betting into a $50k pool (what you expect the final pool to be) and your game is to bet $100 to try to make $1000 (winning more than 10% I hope), around $40k is paid out to the winners so when you get paid along with the others the payout goes down by about 2.5%.

It has been said that some processes are not scalable. You can see that raising the bet size can reduce your edge significantly.

[osophy_junkie]

That is true, but what I mean is, when they start out they are comparing their coefficients to "clean" odds that are not impacted by them in anyway.

After each meeting as they refine their coefficients, if the syndicate is large enough to be betting odds down to the point they offer no value (according to them), then when you are comparing your coefficients to the public odds your are comparing them to odds that already take those coefficients into account.

I personally don't use coefficients, and doubt many syndicates do currently. But the idea's are the same.

There really isn't a way. The public odds represent what you don't know. The more you influence the public odds, the more it incorporates what you do know. You wouldn't be able to bet the odds down enough to overwhelm it with your own information. Or at least I don't think you'd want to as it would I think increase your potential risk.

[PaceAdvantage]

Hi Thomas,

Your absence along with TrifectaMike is conspicuously missed, but I think I understand why.Yes, I'm sure you don't.

[Cratos]

Yes, I'm sure you don't.
Don't know why you are cynical, but their PMs and emails to me suggest that I do.

It has nothing to do with your site, you or anyone associated with this site. People have the right to associate with whomever they please and in whatever audience.

[Dark Target]

Disappointing he isn't here anymore if that's true.

Would be good to hear from him or Magister Ludi.

[Jeff P]

It seems to me you would want to understand how much your bet will affect the payoff. From there you could adjust your betting parameters to allow for the reduced payoff.

For example if you are betting into a $50k pool (what you expect the final pool to be) and your game is to bet $100 to try to make $1000 (winning more than 10% I hope), around $40k is paid out to the winners so when you get paid along with the others the payout goes down by about 2.5%.

It has been said that some processes are not scalable. You can see that raising the bet size can reduce your edge significantly.Imho, this post is spot on.

Not only is the player tasked with developing a +EV model that has to perform well going forward in time after its development - but before said model can be implemented effectively in a parimutuel world- the player also needs to be armed with reasonably accurate estimates of final pool sizes (which vary not only by track code but by by time of year, day of week, etc.)


When any of the larger syndicates start out they will either combine the log of the public's odds with their own, or combine the public odds with their independent variables to get their coefficients.

Initially when doing that process, the public's odds will be completely unaffected by their independent variables giving very good estimates.

Once the model is complete and they start betting into the pool, assuming the public odds are now significantly altered by their independent variables for most of the way through betting to the point of basically controlling the final odds, how would they obtain accurate coefficients going forward, given that now their independent variables presumably add nothing to the public estimates, because the public odds are now essentially their odds?

The following table shows two rows or records (for a mythical race) that illustrates how I handle the bolded text from the question posed by Dark Target at the top of this thread:

RaceDate Track Race Runner TotalWinPool Rake AmtWagered Odds Log(Odds) OurWager OddsAdj Log(OddsAdj)
---------- ----- ---- ------ ------------ ---- ---------- ---- -------- -------- ------- -----------
02-28-2015 AQU 3 1 100,000.00 0.16 7,852.00 9.65 2.266958 1,750.00 12.80 2.549445
02-28-2015 AQU 3 2 100,000.00 0.16 17,893.00 3.65 1.294727 0.00 3.65 1.294727Note that the table contains columns that enable not only recording of pool size, amt bet by the public, public odds, and log(PublicOdds) for each runner... but also allow for recording of the player's bet, adjusted odds where the player's bet has been backed out of the pool, and log(AdjustedOdds) where the player's bet has been backed out of the pool.

The added info (OurWager, OddsAdj, and Log(OddsAdj) fields) enables calculation of coefficients both within the context of where the final pool ended up - and where final pool would have ended up had the player not bet at all.


-jp

.

[Cratos]

Disappointing he isn't here anymore if that's true.

Would be good to hear from him or Magister Ludi.

I agree, but I am a strong supporter of individual rights and freedom of speech.

Therefore if a former poster no longer want to post or will post only sparingly that is his/her right and I vehemently support their right to do so.

Also we will move on to new ventures and interests and because of time constraints we have to be more judicial with our time and therefore we will reduce our postings either by the number of posts or the forums we post too.

[Dark Target]

Hi Jeff

Thanks for the reply. I'm not concerned with how "my" bet will effect the odds from a betting perspective, i'm already aware of that.

I'll try and phrase it another way. For simplicity sake, lets say my model has 2 independent variables, the public's odds and my odds. The data set that I am training it on is for the year 2014, I never bet into that pool at all during that time, so the public odds are in no way influenced by my odds.

I start betting into that pool on 1 Jan 2015, my model is so good that any runner i bet, i bet it down to the point that I think they are no longer value.

My coefficients need to be updated continuously. The 2 independent variables now are the public odds (which are now public odds + my odds), and my odds. Presumably now when I add my odds to the public odds, my odds wont add any value, because the final public odds are basically whatever I bet them down to.

How do I work out the coefficient of the public odds without the effect my odds have had on them, or can it simply not be done?

[Magister Ludi]

How do I work out the coefficient of the public odds without the effect my odds have had on them, or can it simply not be done?

You may be able to estimate the effect from comparing the pari-mutuel odds with fixed odds betting exchanges.

[Jeff P]

Hi Jeff

Thanks for the reply. I'm not concerned with how "my" bet will effect the odds from a betting perspective, i'm already aware of that.

I'll try and phrase it another way. For simplicity sake, lets say my model has 2 independent variables, the public's odds and my odds. The data set that I am training it on is for the year 2014, I never bet into that pool at all during that time, so the public odds are in no way influenced by my odds.

I start betting into that pool on 1 Jan 2015, my model is so good that any runner i bet, i bet it down to the point that I think they are no longer value.

My coefficients need to be updated continuously. The 2 independent variables now are the public odds (which are now public odds + my odds), and my odds. Presumably now when I add my odds to the public odds, my odds wont add any value, because the final public odds are basically whatever I bet them down to.

How do I work out the coefficient of the public odds without the effect my odds have had on them, or can it simply not be done?
See the second part of my previous post.

In addition to recording PublicOdds and Log(PublicOdds) - I suggested recording the following for each runner:

a. Actual amount wagered by the player.

b. AdjustedOdds. This is what public odds would have been had you not bet at all. (Same thing as public odds with your bet backed out of the pool.)

c. Log(AdjustedOdds) - This is what Log(PublicOdds) would have been had you not bet at all. (Same thing as Log(PublicOdds) with your bet backed out of the pool.)

In other words, create two new data points:

Use pool amount, takeout rate, and the amount you wagered (if anything) to recalculate the odds back to what they would have been without your bet in the pool. Call this AdjustedOdds just to give it a name.

Then calculate Log(AdjustedOdds.)

Record both as new data points.

Now you have the ability to evaluate AdjustedOdds and Log(AdjustedOdds) going forward as you continuously update the model.

If you think about it, the two new data points, AdjustedOdds and Log(AdjustedOdds) are essentially the same thing as PublicOdds and Log(PublicOdds) observed during the model's initial development phase.

Hope I managed to explain most of that in a way that makes sense.


-jp

.

[Dark Target]

Hi Jeff - many thanks, i'm not sure how I missed that the first time around.

Much appreciated.