combining losing factors into profits (?)

[formula_2002]

Has anyone ever read about combining independent losing factors into a compound that will result in profits ?

If so, I would like to read about it, providing it has nothing to do about horse race betting.

I’m looking for some works independent of all those books and papers about horse racing.Books and papers that we all have read.

[GameTheory]

Most work along these lines would have to do with predictiveness rather than turning monetary losing propositions into winning ones. The idea that we want to find factors that are not only predictive, but also under utilized by the public at large, seems to me pretty much a pari-mutuel specific problem.

A have a mountain of papers on the former (prediction), but everything on the latter has to do with horse racing...

[formula_2002]

GT well stated. So let me apply my thoughts to your statement.

Predictive ness
There is much empirical data that supports the fact that the pari-mutuel betting public establishes accurate winning odds. There is a short price bias, but that skewing is understood.
In other words they can predict the out come of a race.

Utility.
Because the pari-mutuel bettor has the ability to accurately predict the out come, then it would appear to me that he has fully and completely utilized his information.
Now I ‘ll grant that there is a variance in the odds , perhaps cause by the lack of proper utility, more likely perhaps to other causes on variance.
The question I have to ask is, is that variance sufficient to overcome the track take-out and will it allow for long term profits?

Perhaps, in oder to keep the question out side of the horse racing community the question should be;
“Is it possible to be predictive without utility?”

[cj]

:confused: :confused: :confused: :confused: :confused:

CJ

[formula_2002]

that 's very very funny and also well said (lol)

[ranchwest]

the pari-mutuel bettor has the ability to accurately predict the out come

That is not an accurate statement.

An accurate statement would be that the pari-mutuel public has the ability to accurately predict the probability of the outcome.

[formula_2002]

I fully agree....you said it correctly....I was saying it based on what I felt was the context of the note...regarless


"An accurate statement would be that the pari-mutuel public has the ability to accurately predict the probability of the outcome"

is 100 % correct...


Joe M

[GameTheory]

Every time I tell the following to anyone, they respond with either a blank stare, or if on the internet, no response at all. I'll try it again here. I'm going to spell everything out so it is clear, although there is nothing complicated about it.

As you know, an odds line is the same thing as a set of probabilities that represent the chances of each horse to win the race according to you, the public, or whoever is determining the line in question. (i.e. 2/1 = 33% chance of winning, expressed as a probability is .33) The probabilities in a race must all add up to be exactly 1.0. With the public's odds, you have to renormalize them to add up to 1.0 because of the takeout. For the rest of this discussion we'll assume we've already done that.

Horse racing authors, when discussing the creation of personal odds lines, often say things like, "Imagine this race was run 100 times. How many times would each horse win? The percentages are your probabilities." That view takes some heat around here from the people that think the same horse would win every time, i.e. that the chance of the eventual winner winning was really 100%, we just didn't know it before the race.

Lindsay pointed out a week ago or so that another way to look at this whole odds line thing is to think of the probabilities as not representing the chances of the horses to win, but the chances of you being correct that this is the one horse that is actually going to win the race. (i.e. over the next 100 races like this, I expect to be right this amount of the time)

Whatever view you take doesn't really matter -- you're still assigning probabilities to horses, and they've still got to add up to 1.0 to make a complete line.

Now then, let's talk about the public's line, and its accuracy. (For those wondering if there will be a point to this discussion forthcoming, the seed of it is here.) We always hear how deadly accurate the public is at picking the winners, and that while errors are made by the public in individual races (which is what makes it possible to profit), in the long run over a sample of races the public odds represent the "true" chances of the horses. This can plainly be seen by the fact that the horses win in almost exact accordance with the probabilities as determined by the public. If you look at 1000 horses from 1000 races, each of which was assigned a probability of .25 by the public, you'll find that almost exactly 250 of them (25%) actually did win. How nice. So that means that the public odds are in fact the "true" odds, right? The "correct" line, in the long run, is the public one, right?

This view seems to be near universal, and that's what I can never figure out because it is plainly incorrect. Are the public's odds extremely accurate? Yes. Does that mean that they represent the "true" chances of the horses? NOT AT ALL.

There are a number of reasons why this is a fallacy. These seem obvious, but since some people seem to credit the public odds with magical powers, I'll list some of them:

1) The "true" chances of each horse CANNOT BE KNOWN. The public odds are just a mass opinion.

2) The probabilities set by the public are set independently of the running of the race. They do not affect it. In a non-betting race where there are no odds, do none of the horses have a chance to win?

3) You can set an infinite number of odds lines for a single race, all completely different, but all with EXACTLY THE SAME ACCURACY.

Wait a minute, that last one can't be true, can it? Well, actually, no. But it is true using the means by which accuracy is normally measured. There is in fact only one truly accurate line for a single race -- that in which you assign a probability of 1.0 to the eventual winner, and probabilities of 0.0 to the rest of horses.

But normally people don't determine how accurate their probabilities are by looking at a single race. They look at a large sample of races and note the differences between the expected # of winners as determined by small ranges of probabilities with the actual # of winners that came from each group. For instance, you might find that in 100 races you assigned 100 horses probabilities in the range of .20 - .25 and of those horses, 23 of them actually won. If the rest of the ranges showed such strong correspondence, you would conclude that your probabilites over time are quite accurate. And you'd be right. But if you then decided that since your probabilities are so darn accurate, you have the best lines that you could possibly make, you'd be dead wrong. But this is exactly the procedure by which people determine that the public's odds are "true" -- they are almost perfectly accurate, therefore how could they possibly be better?

The thing is, accuracy as defined above is the wrong way to judge an odds line. We should be judging the information content of the line, not how well the expected matches up with the actual (they are correlated, but they are not the same).

Here's a way to generate a 100% accurate (as defined above) line for any race: divide 1.0 by the number of horses in the race. Every horse gets an equal chance in every race. The observed matches up with the expected even better than the public line, in fact perfectly. (Of the number of horses assigned a probability of .10, exactly 10% of them will win.) But I hope no one would now claim that every horse has an equal chance in every race, that those represent the "true" odds. But that is exactly what people claim about the public odds. The true probabilities cannot be known. The information content of a line that gives every horse an equal chance is 0. It gives you *no* information about this race. So it is not just the accuracy that counts, it is the "spread" of the probabilities -- the higher the probability we assigned the winner, the more information the line contained (we were "less surprised" by the result than if we had assigned a lower probability.)

So how do we determine the value of a set of probabilities? What do I mean exactly by "information content"? What is our true benchmark?

Our true benchmark is the line we could have assigned to the race if we were being really bold -- the one where we give 1.0 to one horse and 0.0 to the rest of the horses. The deviance from that perfect line for each race is how we determine how valuable our line, the public's line, any line, actually is.

All this basically boils down to the question: how many winners did you put on top? The public picks roughly 33% winners on top, very good indeed. But their line is just one of an infinite number that could have been assigned.

What about horses we ranked 2 & up? We assigned probabilities to them too -- don't we need to check to see how well they did? Not really. If you make a line where you assign a probability to each horse and you make sure it always sums up to 1.0, then maximizing the number of winners ranked #1 will tend to naturally put all the ranks in line.

That's the common sense quick version to evaluate things. The precise mathematical way to measure information content comes, strangely enough, from the field of "information theory". The best way to evaluate an odds line is to give it the following score:

log(odds of winner + 1) / log(2)

"odds of winner" is the odds assigned by you, or whatever line in question. The formula can also be expressed as:

log( 1/probability assigned to winner ) / log(2)

Example:

log( 3/1 odds on winner + 1 ) / log(2) = 2
log( 1 / .25 ) / log(2) = 2

Lower is better. The base of the log is unimportant since the ratio is always the same.

Sum up this score for your lines over a sample of races and take an average. Minimizing that number is your goal when trying to improve your lines. It is the mathematical expression of the amount of "surprise", according to your line, that occured when the winner won. That's why lower is better -- minimizing surprise maximizes "information" (non-surprise).

So where does this leave the public odds line? Hopefully you can see that it is just another line -- the thing that is important about the public probabilities are that they set the prices that will be paid to the winning bettors. In fact, they are the prices.

But, please let's stop worshipping their "accuracy".

[formula_2002]

GT, I just printed out your reply. It looks interesting. I will comment on it as time allows.

In the mean time I generated a report on over 116,000 horses which I think will make a case for the public’s accuracy in predicting the winning horse's odds in 73% of the races...

In the other 27% of the races the publics estimate may be considered volatile

[formula_2002]

GT, I reviewed you post. I went to the mirror and practiced a blank stare…
I managed to get through the entire post, even after absorbing,

“Are the public’s odds extremely accurate? Yes. Does that mean that they represent the ‘true’ chances of the winning horses? NOT AT ALL”

Your wisdom is lost on me..

Joe M

[Derek2U]

GT ... I also printed out your math reply & I will re-read it and
I'm sure I won't understand it better but I will consult with
some1 who learned what a log really is. Anyways, in looking
at races as I do, and who might be the eventual winner, I start
this way: Who has the public picked as #1 and #2 --- FAV &
CO-FAV. (Those I note on my DRF)... Then I assign ZERO Pr to
as many entrants as I see fit. Usually, this is almost always
1/2 the field. I then focus on how I see the FAV & 2ndFAV.
Many, many times I toss out either one --- and I am rarely wrong.
I would guess that I do this for about 90% of all races --- and
I toss out the winner (assign a ZERO Pr to it) 10% of the times.

[GameTheory]

Originally posted by formula_2002
GT, I reviewed you post. I went to the mirror and practiced a blank stare…
I managed to get through the entire post, even after absorbing,

“Are the public’s odds extremely accurate? Yes. Does that mean that they represent the ‘true’ chances of the winning horses? NOT AT ALL”

Your wisdom is lost on me..


Well, I guess I haven't lost my touch then.

Here is a shorter version of the one particular point you are hitting on:

Imagine the public bet an equal amount of money on every horse. Would their odds still be extremely accurate (accuracy meaning expected winners = actual winners)? Yes, even more so than they are now. Does that mean that the "true" chances of the horses are all exactly equal? (At this point you should be saying, "Of course not.")

The public is very good at picking winners. Agreed. But they could pick many less winners (betting equally on every horse) with even HIGHER accuracy. They could pick even more winners with the same accuracy. Expected = Actual IS NOT THE THING. IT IS EASY TO ACCOMPLISH THAT (just give every horse an equal chance in every race).

I have a hard time talking about this because I actually don't know what I'm arguing against. It seems to be some sort of mysticism. You reply with things like, "I have a report that shows they're very good indeed at predicting the winner." My point on a simpler level is SO WHAT? It is as if you are staring at me and saying, "What do you mean, so what? THEY ARE *REALLY* GOOD at predicting the winner!" So what?

All that means is that the prices on winners tend to lowish most of the time. That is all it means.

[formula_2002]

http://www.flatstats.co.uk/articles/impact_values.html

[GameTheory]

Originally posted by Derek2U
GT ... I also printed out your math reply & I will re-read it and
I'm sure I won't understand it better but I will consult with
some1 who learned what a log really is.


I don't bother with the explanation of the math since the post was so long already but understanding of logs is not neccessary. Just push the button on your calculator or use the log() or ln() function in any programming language.

Information in its simplest form is measured in bits (1's & 0's). The more bits you need, the more information in the "message" you are encoding. In this case, the message is the extra information we need to add to our line after we know who won the race in order to reflect that fact. The shorter the message, the less suprised we were when the winner won (because we gave the winner a higher probability), and the less bits we need to encode that message. Information theory is the brainchild of Claude Shannon (who recently died, by the way) of the old Bell Labs. The Kelly criterion also comes out of information theory. (Kelly worked with Shannon at Bell.)

[sjk]

GameTheory,

Try again- I have been reading the forum for quite some time. I find your post useful enough to register to thank you. I make an odds line for pretty much all the races and have been looking at making some small changes in my program. You suggest a much better way to testing the effectiveness of the changes. Thanks.

[Rick]

What you really want to do when you're establishing a line is to come up with an independent estimate of odds (or probabilities). Even though the public odds estimate may be more accurate than yours, if your line is relatively independent of it they can always be combined to produce a more accurate estimate than either one used alone. If your line includes mostly things that are already incorporated into the public odds, it will be of no value whatsoever. For example, adjusted morning line odds is usually a better estimate than any line you can calculate. But, it's entirely incorporated into actual odds and will add no value to an estimate based on actual odds.

So, if you really want to find out if your line has any value at all, combine it with actual odds in a regression equation. If the odds line variable is statistically significant you may have something. If not, better get back to the drawing board because even if you're having short-term success, I'll guarantee you that it won't last.

[sjk]

Rick

I hope I have understood your suggestion correctly. I used the regression tool in Excel with 1(won) or 0(lost) as the y variable and my probablity line along with the actual toteboard probablilties as the x variables (65533 starters). The coefficient of my probabilities is .299 and the coefficient of the actual toteboard line is .241. I interpret this as a positive result. Should I be looking at any of the other output.

It would seem that I could use this sort of analysis to find weaknesses in my line-making program by trying to see which line is more effective in responding to changes in variables which the public is using making the odds. Any suggestions on how to do this and what variables to include? Thanks.

[GameTheory]

If I may,

That's basically what you want to do. Logistic regression will work best. Then you use the resulting equation to create a new blended line. Of course, that has to be done in real time with the actual odds, although it might work ok with the morning line as well.

Now give the public line & your new blended line each a "score" as I described earlier over a sample of races (a different sample than you used for the regression). If the score for the blended line is lower than the public line, there is a very good chance of making money betting overlays off that line. You can also score your original line, of course, but it is much less likely to be superior to the public line...

[kenwoodall]

In the book "a day at the races" northern Ca official pricemaker Richard Summers tells how the makes the morning line for each race in about 20 minutes- and it is very accurate because the bettors usually bet to adjust the tote to the ml. They think a horse 2-1 on the ml and 4-1 on the tote neds to be bet down! a self-fullfilling prophesy! Then the first day at the fairs he makes a "favorite" 8-1 so suckers think they hit a long shot!
2001 Stockton first day Every mule, appy, thoroughbred paid at least $18.00!
You have to know the rules, then how and when the rules are waived!! You have to beat the odds by 25% to profit consistently!

[Rick]

GT,

Thanks for the input.

sjk,

Most regression software also gives a significance level for each variable which should be 5% or less. That tells you a lot more than just looking at the coefficients But with the values you stated, it sounds like you probably have something significant anyway since I'd expect the coefficient for your odds line to be close to zero otherwise. GT's advice sounds good to me, except that I wouldn't use morning line odds since they seem to most often represent "dumb money" and you don't want that included in your line. I wouldn't use this procedure to calculate a super accurate odds line because of the logistical difficulties involved in incorporating late odds though. I'd just use it to prove significance and decide whether to include new variables. If you could include actual odds, the combined result would just be closer to the actual odds anyway, so you can just use your line alone and demand a bigger overlay.

kenwoodall,

What factors does he use in creating his morning line?

[sjk]

Rick

When I bet online, I have the odds line in the computer anyway and download the supertote board to identify overlays. With that set up, it is no big deal to make a linear combination of probabilities in real time. It seemed to me that to take a combination of my line and the actual odds really amounted to setting a higher hurdle for the overlay % (I look for 75% overlay anyway).

I tried to use the combo line on certain types of horses where my results appeared weak (2nd time starters) but was not pleased with the results. Instead, I plan to adjust upward the overlay % required on these horses; I will get fewer plays, but it will not push me into possible bad plays on other horses in the race.

[Rick]

sjk,

Sounds like a good plan. If you can easily incorportate actual odds you might try looking at the differences at various points in time. Some say that 3 minute odds are much more accurate than 10 minute odds for example.

[kenwoodall]

ml= head to head, finishes, beyers, opinion.

[GameTheory]

I would log the results into two categories -- those races where your top choice & the public's are the same and those where they are not.

In the first category, the best thing may be to only consider betting on that top choice (the favorite), and only if it is still an overlay (a low-priced one). This is the "do not bet against a legitimate favorite" rule, no matter if you have apparent overlays on other horses. This will save you money in most cases (all of this depends on the specific character of your betting line and the factors that go into it, so do testing).

The second category is where you should be betting agressively (on possibly two or more horses) against the favorite and making your real money. I would expect to find 3 or 4 false favorites on a card....

[Rick]

GT,

I agree. I get about 4 false favorites on average every day. The best I can do from the rest is to break even if I select only those from the rest that have very strong factors going for them (the legitmate favorites). I wish I could do better than break even on them because they win a really high % of the time (like maybe 36%). If anyone has any suggestions here, I'd appreciate them.

[GameTheory]

Sometimes the place & show ROI's are a few percent better on low priced horses in the break-even range. Of course, they hit with really high rates.

I would also look into putting the favorite on top of your 2nd or 3rd choices in the exacta....

[Derek2U]

Just a remark .... I still think that all analysis must include a hard look at the favorite ... and for me, the co-fav when its a low-price also. I haven't kept any stats, but when I rate the Fav as ZERO, I'm rarely, rarely wrong; I won't even consider it On Top in any wager. On the other hand, when I think the FAV is a LOCK to Win, I almost NEVER bet the race, except these days, it will end
up my P4 lock for that race. I'd like to hear how everyone deals
with those less than 2:1 horses.

[sjk]

I looked at my theoretical results during the last 6 months or so (29388 dirt races since 6/4/2002). I had my computer handicap all of the races since that date using previously known information (there is a bit of leakage of post date info which I can't eliminate but don't consider significant).

My rate of return on the races where I agreed with the public on the favorite (18779 out of 29388 races) was nearly identical with the races where I disagreed with the public on the favorite.

I tend to look at the process as an exercise in patience where I bet overlays and may have to wait a long time for the horses to get the job done. (Watching the races does provide excitement win or lose.) Favorites get beat both because all of the available information makes them false favorites and because of bad racing luck. Thers is nothing wrong with betting against the horse you consider the best if in the long run the bets pan out.

I bet mainly exactas and I do bet a lot of tickets with favorites on top and overlays underneath.

[Rick]

GT,

Exactas are profitable on those races but I don't bet them except for recreational purposes. The bet size would have to be much smaller so I don't bother with my real bankroll. Second choices to win also show a profit but at a much lower win % which again would call for much smaller bets. I haven't looked at place bets yet but that might be the answer.

[Rick]

Derek,

Short priced horses may be an overlay when they don't have the best last race speed rating. You should keep records though. I've found that it's really dangerous to rely only on my subjective feelings about how well I've done.

[Aussieplayer]

Hey guys,

Good thread. In the past (and over more than just the one rating method), I have tested various analyis processes (re: taking into account the public's odds) that are perhaps a little more "practical" for some (read, easier, lol), than some of the methods talked about here. I'd like your (esp. Rick & GT's) thoughts.

I've looked at all the various combos of my actual "raw" odds vs. public's actual odds, rank vs rank, actual vs rank, rank vs actual etc. etc. etc. etc.

By far, the analysis that has consistently shown to be the best (for me), is simply degree of overlay.

After digesting all the thoughts of this thread, and similar comments in others, I can't help but wonder if (in all reality), this method "takes into account" the public odds (which is what we want to do), just as "much" as any fancy mathematical analysis? That's not meant to sound like a downer on the maths stuff BTW.

It just strikes me that the crude method I've been using to analyse odds lines really DOES account for the public - and is very easy!!

Hope that helps someone.

Cheers
AP

[GameTheory]

If it works, it works.

The idea of mathematically combining a line with the public's is to produce a line more accurate than either of them. But if you just use your line and insist upon a certain degree of "overlayness" and it works, then it works.

But you might find that certain odds ranges require a greater margin than others due to bias in your personal line, and the blending process helps smooth that out and you can just bet any overlay with the required edge with confidence. But it is more trouble to do all that true, and not necessarily worth the effort.

[Aussieplayer]

Hi GT,

You said, "But you might find that certain odds ranges require a greater margin than others due to bias in your personal line, and the blending process helps smooth that out and you can just bet any overlay with the required edge with confidence. But it is more trouble to do all that true, and not necessarily worth the effort."

.....This is exactly what I was trying to lay out, lol

What I am proposing to you, is that this method should/could/would work as well, or is as relevant, (perhaps), as more mathematical tests. What thinks you?

Perhaps it would have been clearer with an example.
You might find you can bet your top rated between 1 and 2 times your assessed price. Under that and it loses, over 2 times and you have....a false overlay!!
Of course you can then break this down into various race profiles, betting profiles (where the fav is in your contenders), your actual rated odds etc. But that's another story.

Cheers
AP

[GameTheory]

Obviously you need some amount of margin from (odds greater than) your line to have an edge. But you might find that you need a greater margin at different odds.

For instance, the public says 3-1, you say 2-1, you bet. You find this is a profitable situation.

Now, the public says 6-1, you say 7/2, and you find that in this situation you only break even or even lose money. What's going on? You may need 5/2 or better on your line when the public says 6/1.

A mathematical blending can smooth that out so a 20% edge (for example) always means a 20% edge no matter what the actual odds.

I have a feeling I'm making this more confusing...

[Aussieplayer]

....GT,
I understand. I think. You seem to be saying that I should go a step further, and actually change my line. In your example, when my line has 7/2 and public says 6/1, my line should be changed to 5/2. But why? What difference is changing my line going to do?

I would rather know that when my top choice is rated (by me) at 7/2 and the public says 6/1 (or more) that I have a "no play."
If you see what I mean.......

[GameTheory]

You should change your line the opposite way.

The odds on your line should move towards the public (in general), but the amount it should move is dependent on both the odds you initially assign and the the public odds. Your 7/2 might become a 5/1 and not enough edge to justify a play. If it started out at 5/2, it might become 4/1 with a still acceptable edge. How much edge you require is up to you but you have to leave some room for error.

I think you will find the greater the difference between the your line and the public's, the greater your line is in error (when the public odds are higher than yours) and the overlay is never quite as big as you think it is. Using the regression technique to blend the lines removes much of this error, and then an overlay is an overlay, period.

These ideas are all straight from Benter, by the way.

[kenwoodall]

Of course, overly complicated handicappong can cause loss of focus! My show sucess rate is high enough not to worry about odds, and my win picks are very strong picks or overlays with positives not based on speed or class. Too many bettors lose by betting too heavily ob speed and class, causing a personal underlay.
Aussieplayer- email me at kensav@webtv.net. University of Sydney has a great website on equine research and Australian track analysis!

[Rick]

AP,

The only reason to combine your line with actual odds is if you want a more accurate line than either one alone. That would allow you to really say you had a 20% overlay when the probability ratios showed that. But if your willing to ask for say a 50% overlay in probabilities and realize that your actual overlay won't be that large, then just use your calculated odds alone. I think we only got off on that tangent because there has been so much discussion of what the most accurate line would be.

I'm not really for complicating things at all. I don't even compare to actual odds, partly because final odds are not available before the race and partly because of convenience. I'm not sure what you guys have available there. You must have some sort of morning line odds or "starting price" or something. Do you have bookies as well as parimutuel betting?

As far as testing for statistical significance goes, it's helpful to find out as soon as possible, with a smaller sample, how much faith you should put in your method still working in the future. If you have a very large sample with a good ROI it's probably significant even if you can't show exactly what level of confidence to assign to it. I've tested several samples for my method in the past but I won't be bothering to repeat that after thousands of races and almost the same ROI that I used to prove significance before.

[Handy Cap]

2 things.....

number 1, great great thread...I believe maybe the best I have read on PA!

number 2, So let me get this straight GT...If your odds say 2-1...and the public says 7-1 or greater perhaps...then theres a good chance that you have given the animal too much credit, and should sway his odds upward?? Am I on the same page here?

[formula_2002]

It might be of interest for some to view an excel table I have posted on my web page. You should be able to print it out

"How well the public does in determining the true winning odds in horse racing for over 91,000 uncoupled horses"

The table list the stats for segmented odds ranges .10 to 20-1.

I also indicate my odds line in relationship to the public's

Pehaps some one can offer advice on how I can get and edge, using my odds line.

Web page is my personal page, and is not a commercial site.

Thanks
Joe M

[GameTheory]

Originally posted by Handy Cap

number 2, So let me get this straight GT...If your odds say 2-1...and the public says 7-1 or greater perhaps...then theres a good chance that you have given the animal too much credit, and should sway his odds upward?? Am I on the same page here?

Yeah, exactly.

[cj]

What difference would it make to adjust your odds upward? Regardless, the horse would still be considered an overlay up until you made him the same odds as the public. It this point, why make a line? I'm saying adjusting your odds line is a waste of time, like Rick said, just demand a higher % overlay.

CJ

[Dave Schwartz]

CJ,

The concept is well-grounded in reality. We do the same thing.

COnsider this: You have 2 horses that you both give a 25% probability to (or 3/1 if you prefer). One of them goes off at even money and the other at 20/1. Do you really believe both of those horses have the same probability?

So, in my view, the issue becomes not, "Should I move my line?" but "How MUCH should I move my line?"

Regards,
Dave Schwartz

[GR1@HTR]

Originally posted by GameTheory
As you know, an odds line is the same thing as a set of probabilities that represent the chances of each horse to win the race according to you, the public, or whoever is determining the line in question. (i.e. 2/1 = 33% chance of winning, expressed as a probability is .33) The probabilities in a race must all add up to be exactly 1.0. With the public's odds, you have to renormalize them to add up to 1.0 because of the takeout. For the rest of this discussion we'll assume we've already done that.


For those not familiar with our lovely tote system, the total probabilty of all the horses winning is not 100%. The total of the percentages is about 125%.

Example:
FG 2/9/03 Race 1 6f dirt for 4 YO & Up Fillies & Mares

C25000 Track fast Time 22.99 47.20 59.80 113.16 Weather cloudy

-----------------------------------------------------------------------------
sc horse pp st 2f 4f 5f 6f odds
-----------------------------------------------------------------------------
4 Ice Queen 4 6 3-2 2-0.1 1-0 1-0 2.9
2 Pearly Starnet 2 4 5-2.5 4-3 4-2.5 2-0.05 14.8
7 Plush 7 1 6-4 5-4 5-5 3-3.5 5.6
6 Zarb's Love 6 2 2-1 1-0 2-0.1 4-4 4.1
1 Sacred Gem 1 5 4-2.5 6-5.5 6-6.5 5-4.25 1.2
5 Just Robin 5 3 1-0 3-1 3-1 6-7.25 22.7
3 Hilga 3 7 7-5 7-7.5 7-9.5 7-7.5 17.4
-----------------------------------------------------------------------------

4 Ice Queen.................. $7.80 $4.20 $3.40
2 Pearly Starnet............. $9.80 $5.40
7 Plush...................... $5.80

EX 4-2 $86.20
QU 2-4 $42.20
TF 4-2-7 $397.60

Horse#...........Odds........Probablity based on final odds
#4.................2.9.............2564
#2.................14.8...........0633
#7..................5.6...........1515
#6..................4.1...........1961
#1..................1.2...........4545
#5.................22.7..........0422
#3.................17.4..........0543

The sum of the above probabilities is 122 for the above example vs the assumption that the total of the odds is 100 or 100%

[GameTheory]

Originally posted by cjmilkowski
What difference would it make to adjust your odds upward? Regardless, the horse would still be considered an overlay up until you made him the same odds as the public. It this point, why make a line? I'm saying adjusting your odds line is a waste of time, like Rick said, just demand a higher % overlay.
CJ

I never said, "This is something everyone should do." It is something to do if you want or require the most accurate probabilities possible. Most people don't care (not that they necessarily should) about the degree to which they have an edge, only that they have one.

But if you are varying your bet size, for instance, with the amount of the perceived edge, you'd better take care to get it as close as possible. If all you are interested in is, "Do I bet this $20 on this horse or don't I?" then you're right, it is probably a waste of time and you just need to demand a decent margin to assure value.

It is also a technique which lets you know how much "value-added" information is in your own line. If after making a blended line via regression or similar, you find your line to more or less disappear (i.e. the blended line is no different than the public line), then you know your line is crap and is adding no information not already accounted for by the public odds.

Whether or not any of these ideas are of any worth to your particular approach is up to you...

[formula_2002]

GR!, thanks for the post,
Ice Queen was one of my 4 winning picks on Sunday....out of 31 picks...what a crushing weekend....

My line on her was 2.75-1 she went off at 2.90-1

joe m

[cj]

Originally posted by Dave Schwartz
CJ,

The concept is well-grounded in reality. We do the same thing.

COnsider this: You have 2 horses that you both give a 25% probability to (or 3/1 if you prefer). One of them goes off at even money and the other at 20/1. Do you really believe both of those horses have the same probability?

So, in my view, the issue becomes not, "Should I move my line?" but "How MUCH should I move my line?"

Regards,
Dave Schwartz

I understand what you are saying Dave, but what difference does it make how much you move the line? Unless you are going to move it to that of the public line, you are still going to have the underlay as an underlay and the overlay as an overlay. Isn't the purpose of making a line to determine overlays / underlays? My point is, that won't change through adjustment.

I also see GTs point about tailoring bet size to perceived edge, but the other problem is you don't know the actual public odds until its too late, other than a ballpark guess. I don't think anyone could ever make a line accurate enough to benefit from this type betting anyway.

I could see maybe not betting horses with a wide disparity such as the example you cite, but I feel any adjustment of a personal line towards the public line is not a worthy pursuit. Find out why the big difference (where you are going wrong) if you are not profiting on those types of horses, and if you are, don't change anything.

CJ

[cj]

Originally posted by formula_2002
GR!, thanks for the post,
Ice Queen was one of my 4 winning picks on Sunday....out of 31 picks...what a crushing weekend....

My line on her was 2.75-1 she went off at 2.90-1

joe m

Personally, I wouldn't count this horse as a winner. The tote board would have read 5-2, thus appearing to be an underlay. That's how I do things, for what it's worth. Unless, of course, you are in North Dakota :).

CJ

[Aussieplayer]

Hey guys,

Come back 2 days later & the thread just gets away from you, lol!!

GT - thanks for clarifying. It was your earlier example that had me confused where you said to adjust DOWN - must have been a typo. I know where you're coming from now.

kenwoodall - will email.

Rick & GT - just a caution: Even with a line adjusted to the public (you guys both say, "then you REALLY have an overlay"), Benter cautions that you can't actually say that with 100% certainty. I know you guys didn't mean that, just in case anyone else thought it might be the "pot of gold."

Rick - I'm with you re: not keeping things complicated! It may well be enough to just "keep in mind" that you probably don't really have the advantage that you think.
But then, that's what not betting full-kelly etc. is all about.

Also, Benter agrees with that as well. Don't forget - he simply said that he found that adjusting to the public's odds PRIOR to using your line for betting was in fact BETTER than adjusting to the public's odds from your original line, BUT, that doesn't invalidate doing it the other way around.

Yes we do have a morning line. Several actually, depending on which paper you use, but they're usually pretty similar.
And, I wish I had done my line studies against the morning line, and even my current "average 1 per day" against the morning line. I don't want to have to monitor the tote 5 times a day. I just wanna make the bet in the morning, lol
In the "real world," I think your suggestion of using the morning line is brilliant for a lot of us players.
Yes, we have bookies here. As well as three totes.
The exchanges would like to start-up but will recieve opposition.
I think it makes things a little different when you also have the "search for best price" as part of your betting strategy. I'd like to see the three totes merge into 1 big one though.

Dave - I know you were talking to CJ, but I'll chime in anyway, lol
I am suggesting that (for a lot of players), there is a way to:
(a) do better than just "accepting" you probably have an edge, but not worry about how much etc., BUT -
(b) without having a proggie to take the tote info. and blend the lines.

It's the method I've personally found works real well, and is suited to players who (say) have their "sheets" in front of them when they bet (perhaps like Steve Fierro's), and want to take their game further than just blindly betting on the overlays.

Cheers
AP

[Rick]

AP,

Beyer says you guys have a tougher game to beat than we do and he may very well be right. Less information and more accurate odds probably. But, if you can come up with a "dumb money" consensus somehow you may do better than comparing with actual odds.

I wouldn't consider anyone's line very accurate no matter what they included in it. The upper limit to predictability is probably somewhere in the 40-50% range no matter how much information is included. But we don't need to be all that accurate to make a LOT of money, so I don't really care.

[chickenhead]

had somehow missed this thread up to this point, but in light of some recent threads here I thought I'd give this one a bump.

[formula_2002]

had somehow missed this thread up to this point, but in light of some recent threads here I thought I'd give this one a bump.


Wow, it looked interesting to me too.
Then I saw I started it!!! :)

[sjk]

After all this time and countless hours of work I have still not produced a line which does better (in the information theory sense above) than the public line. I have gradually inched far closer than I was two years ago when this thread was going.

Anyone else have a line beating the public line over the long term?

[formula_2002]

After all this time and countless hours of work I have still not produced a line which does better (in the information theory sense above) than the public line. I have gradually inched far closer than I was two years ago when this thread was going.

Anyone else have a line beating the public line over the long term?


Oh my, I just noticed the date (1/31/03) of my first post..
I thought it was but a month or so ago!!

Hey, I can remember trying to figure odds lines while watching Monday Night Football.....with Howard Cossell!!

[sjk]

Formula,

You were actually pretty positive in this thread. Now does it seem like a long time ago?

[JustMissed]

For those not familiar with our lovely tote system, the total probabilty of all the horses winning is not 100%. The total of the percentages is about 125%.


I would suggest that maybe before we get too excited about "Blending with The Public Odds", we adjust the fair odds for the track takeout so we are comparing Apples w/ Apples and not Apples w/ Oranges.

JM ;)

"PARI-MUTUEL ODDS: Odds on each horse are inversely proportional to the actual money wagered on each horse. For example, consider a four horse field with 50 $2 wagers or $100 total. Assume:



Proportion Pari-mutuel Odds
5 $2 wagers on horse A $ 10 10/100 = 10% 1/.10 - 1 = 9/1
10 $2 wagers on horse B $ 20 20/100 = 20% 1/.20 - 1 = 4/1
15 $2 wagers on horse C $ 30 30/100 = 30% 1/.30 - 1 = 2.3/1
20 $2 wagers on horse D $ 40 40/100 = 40% 1/.40 - 1 = 1.5/1
Total $100 100%


In the above example, horse A with 10% of the win pool is a 9/1 shot and should pay $20 to win, including return of the wager. Horse B with 20% of the win pool is a 4/1 shot and should pay $10 to win, including return of the wager. Horse C with 30% of the win pool is a 2.3/1 shot and should pay $6.60 to win, including return of the wager. Horse D with 40% of the win pool is a 3/2 shot and should pay $5 to win, including return of the wager.

TRACK ODDS OR TOTE ODDS: Pari-Mutuel odds adjusted "downward" by the track's take-out. For example, assume a track take-out of 20%. Total wagered = $100, total returned to betting public = $80, then:

Horse A with 10% of the win pool will pay 7 to 1 or $16: [(1 - Take-Out % / % of Win Handle) - 1] = [(1-.2 /.1) - 1] That is, 5 winners each receive $16, total returned to betting public = $80;

Horse B with 20% of the win pool will pay 3 to 1 or $8: [(1-.2 /.2) - 1] That is, 10 winners each receive $8, total returned to betting public = $80;

Horse C with 30% of the win pool will pay 1.6 to 1 = 8/5 or $5.20: [(1-.2 /.3) - 1] That is, 15 winners each receive $5.20, total returned to betting public = $78 ($2 breakage to track);

Horse D with 40% of the win pool will pay 1 to 1 or $4: [(1 - .2 /.4) - 1] That is, 20 winners each receive $4, total returned to betting public = $80."

[JustMissed]

"If you want to calculate the exact win odds on a horse, you will need to use a couple of other figures from the toteboard: total win pool and amount bet on that horse. The total pool is all the money bet on all the horses to win, but it is NOT the amount that will actually be paid out to the holders of winning tickets. Before the track pays off the winners they deduct the "take" which is generally between 14%-20% and is different for every state. You can look up the take at your track here. This money goes to pay state and local taxes, purse money for the horsemen, expenses at the track, and the track's profit. To calculate the exact odds on your horse, just subtract the take from the total pool then subtract the amount bet on your horse to give you the amount of cash to be paid out. Divide that figure by the amount bet on your horse to get the exact odds. This figure will always be rounded off to the nearest dime (usually) or nickel as mentioned above before the payoffs are calculated. Again, those who want to know more about breakage and how it affects your payoffs should read this article. Here is a simple example of win odds calculation:

Total pool: $900
Amount bet on horse #1: $300
Take amount: 15%

$900 - 15% = $765
$765 - $300 = $465
$465 / $300 = $1.55

Round this off to $1.50 for breakage and you get odds of 1.5-1 or 3-2 as it would normally be written which yields a payoff of $5.00 on a $2.00 wager."

JM

Please note that before breakage and takeout, the natural odds as a percentage are 33.3% or 2-1.(300/900).

The tote would show 3-2 or 40%.

Obviously 40% is higher than 33.3% and this is why the tote odds expressed as a percentage are always greater than 100%.

[Kreed]

I read those posts from 2002 and GT still remains brilliant. But that Derek2U ..
wow, now that guy is like Picasso and handsome 2. Damnn. Where is D2U now?

[Kreed]

THe one main problem with all factors is they are very correlated. Nothing
original apart from each other. Has anyone here used math techniques to
purify their factors?

[Jeff P]

posted by Kreed-
THe one main problem with all factors is they are very correlated. Nothing
original apart from each other. Has anyone here used math techniques to
purify their factors?
Someone else on this board once posted (and I think it may have been Michael Nunamaker) "Speed is class and class is speed." I tend to think that there's a lot of truth behind both statements. As speed and pace handicappers, we tend to pay a lot of attention to speed and pace figures. If you think about it, speed and pace figures also measure class. And if you think about it even further, recent speed and pace figures also measure current form. Why else do we tend to put so much emphasis on recent races?

Several years back I set out to try and develop my own comprehensive rating that attempted to roll many many factors into a single number. The goal was to produce a number that could not only be useful for rating horses but for betting them successfully as well. At first, it was like running into a brick wall. All of the factors I was using seemed related to each other in some way. It took me several years of trial and error to come up with a single number that actually works.

I'll be the first one to admit that my number isn't perfect. In fact, there's no such thing as a perfect number. No matter how good a number you produce, there's always some way to improve upon it . After starting such a project, I continually found myself being drawn back to it, tweaking this and testing that, always seeking a way to bring some incremental gain into what I already had. I find this true even today. In fact I spent about two hours today testing something new against races from 2001. Unfortunately for me the new idea didn't pan out and I scrapped it.

Kreed asked if anyone used math techniques to purify their factors. I don't use math techniques so much as I tried to use a little common sense so as to reduce the effect of factor A representing the same thing that factor B does.

I didn't have any success producing a single number until I started thinking about races in very general terms. Essentially, I took a view from a very high level (about 50,000 feet.) What I did was divy up the basic info found in a past performance record into five general categories- each one representing a different school of handicapping thought. Looking at races this way I saw:

1. Early Speed
2. Current Form
3. Raw Ability
4. Class
5. Connections

I set about creating and testing several compound numbers within each separate category. That in itself took a great deal of time and effort. Again, from time to time I find myself going back to look at, tweak, and test what's already in place in the hopes of getting some new incremental gain into the project.

Once I had a compound number that I felt truly represented each separate school of handicapping thought, I spent quite a bit of time experimenting with a weighting algorithm that brought them all together and testing the results of different assigned weights.

One of the more interesting things I discovered is that the public's betting patterns are changing gradually with the passage of time. Using data pre-2001, the whole thing works much better if the early speed compound numbers are assigned a higher weight. Because the public tunes in on different things as time passes, it pays to be aware of what they are currently focusing on and to re-weight accordingly. I would be surprised if the same weights I am using right now would be as effective 20 years from now.

Just realized how long winded this post is getting. Sorry for that. To sum it up, if you are trying to bring many factors together, and are finding that they all seem closely related, try taking a step back. Break them into broad general categories and have another run at it. You just may find that it helps.

[Overlay]

I'm jumping into this thread really late in the game, and my comment may have been touched on before, since I didn't thoroughly digest all the numerous earlier responses. There's a lot of really great information there, but I kept coming back to the original question. On a basic level, isn't "combining independent losing factors into a compound that will result in profits" what the goal of any system is, or should be? And if you have single factors (like the vast majority of performance indicators) that (even if they have a positive correlation with winning probability) independently produce a negative return on investment because of overbetting, isn't the object of system play to combine those separate negative-expectation factors into an integrated model which results in a recurring positive return? However, it seems to me that any such system (even if based on multiple variables, and even if initially profitable) will eventually fail unless it addresses the subject of acceptable odds for either the system's top-rated horses or for all horses in a field, so that you can either lay off choices that have been bet down too low, or else have a sound basis for determining whether other horses in the same race offer value.

[NoDayJob]

In the book "a day at the races" northern Ca official pricemaker Richard Summers tells how the makes the morning line for each race in about 20 minutes- and it is very accurate because the bettors usually bet to adjust the tote to the ml. They think a horse 2-1 on the ml and 4-1 on the tote neds to be bet down! a self-fullfilling prophesy! Then the first day at the fairs he makes a "favorite" 8-1 so suckers think they hit a long shot!
2001 Stockton first day Every mule, appy, thoroughbred paid at least $18.00!
You have to know the rules, then how and when the rules are waived!! You have to beat the odds by 25% to profit consistently!

Rich Somers is one of the best pricemakers in the country. We have been friends for many, many years. What he says, you can take to the bank. A very good handicapper, too. He's done it all.

NDJ

[GameTheory]

After all this time and countless hours of work I have still not produced a line which does better (in the information theory sense above) than the public line. I have gradually inched far closer than I was two years ago when this thread was going.

Anyone else have a line beating the public line over the long term?Using that measure it is very very tough. I have a few more tricks up my sleeve since then, but still use this measure along with some others. I use it as a relative measure now, adjusting for field size:

P = probability assigned to winner
FS = field size

relative info score = 1 - ( (log(1/P)/log(2)) / (log(FS)/log(2)) )

Using it this way, higher is better, and a score of 0 is equivalent to random guessing. A score less than 0 is worse than random.

Luckily, beating the public on this score is not necessary to make money because you can generate value simply by being different, even if you are somewhat less accurate. There is one way easy way to beat the public though, but not with enough magnitude to make any practical difference. And it is proof of the favorite-longshot bias. Just take the public's line, reduce their probability slightly on the longest shot in the field, and increase the probability of the favorite by the same amount. You should be able to create a line that is just a teeny-tiny bit stronger than before. Not that it will do you any good because you can only adjust mechanically this way by a very small amount and still get a better score.

[PaceAdvantage]

Rich Somers is one of the best pricemakers in the country. We have been friends for many, many years. What he says, you can take to the bank. A very good handicapper, too. He's done it all.

NDJ

Holy Dog Doo Batman! A post where you don't mention the quicker slicker pony picker!! What a breath of fresh air!!

Keep it up NDJ, and you just might win "Poster of the Year!"

:eek: ;) :eek:

[kenwoodallpromos]

My point exactly- Somers knows what he is doing and does it well. I just have to make sure I know what he and others in the industry are doing (making the races as unpredictable as possible)!

[NoDayJob]

Holy Dog Doo Batman! A post where you don't mention the quicker slicker pony picker!! What a breath of fresh air!!

Keep it up NDJ, and you just might win "Poster of the Year!"

:eek: ;) :eek:

:D PA, I have my post hole digger and posts at the ready. :D

NDJ