Creating an oddsline

[kgonzales]

I am creating software that implements the Pizzola Hadicapping Magic Method. I use BRIS DRF mutli files for my pp info. What I am trying to do is to create a formula that will predict/create an oddsline that mirrors what the final oddsline should be based on conventional handicapping (speed, form, class, etc.). I am trying to handicap the public to know when I have true and false overlays. I'm thinking that I can just use the Prime Power numbers, but I don't really know know how I would create a properly weighted oddsline based on this number. Any ideas would be greatly appreciated.

Also does anyone know what the points per beaten length for BRIS speed numbers are at different distances. They only list approximations for sprint and route on their website.

Thanks,
Kristian:confused:

[GR1@HTR]

kgonzales,

Best wishes creating a Bris version of Handicapping Magic. Me thinks you can do better do better than the orginal format as presented by ITS.

As far as creating an Oddsline...You first have to determine what the actual win% is for each horse you have. I guess the only way to do that is to create a so called "Power Rating" for each horse and then test the rankings or differentials of each Power Rating. Only then will you have an idea of what your win% is for 1st ranked, 2nd ranked...5th ranked horse. Money Secrets at the Racetrack by B Meadow will also tell you have to create an odds line based upon win%. For a short cut, you can use figline which can be found at www.netcapper.com.

[Dave Schwartz]

KGonzales,

I do not know if it is still the way they do it but I had someone at BRIS explain to me a few years ago how they did it.

Essentially, they look at the average beaten margin between horses at each track and distance. That becomes a "point."

Huh? That's right. Every track and distance has its own unique value of a point and it is (was?) based upon the average margin between horses. And that is ALL horses, not just first and second finishers.

Hope this helps (although I cannot imagine how).

Good luck.

Regards,
Dave Schwartz

[Rick]

kgonzales,

I don't think you should use Prime Power as a way to predict actual odds because it loses less than the average. What you really want to do is predict "dumb money" based on factors that everyone uses. My guess would be speed ratings in last two races, average earnings, and jockey win %. When you've chosen your factors, you can use multiple regression to predict odds. I would use log(odds+1) as the dependent variable because the fit will be better. Also, don't forget that morning line odds (adjusted for scratches) is a very good predictor of actual odds on it's own.

[andicap]

Eric Langjahr actually put together a program purported to predict what the public will do. It might be part of Handicapping Magic software. I know uses a Beyer algorithim as one factor, but not sure what else.

[Tom]

http://www.a1handicapping.com/tsn.html

This link will get you the TSN values, which I think are the same used by BRIS.

[tanda]

Ray,

Are you saying that your conversion of power ratings to an oddsline cannot be done by computer?

If you explain a procedure in the book, then it should be able to be coded.

As to "binary calculation", what the hell does that mean? Are you saying computers can only compute binary calculations? The basic language of code is binary digits ("bits") but a computer can be programmed to perform a wide variety of calculations, in fact, any type of calculations a human can do. If a computer only can calculate binary (and I really do not know what that means), then humans only can as well.

Did you mean linear? Do you even know what you meant other than hawking a probably poor product?

The example you presented is true, but there are algorithms that can be written to deal with it. They can consider not just the rank but the degree of difference among ranks. That has nothing to do with "binary calculation".

But I guess only your supersystem can handle those complex calculations.

It seems like a lot of the new posters are selling something (and some of it smells).

I propose a new rule:

Before anybody can post a web-site address, product name, contact information, etc., they must post (before post-time) a series of 100 consecutive plays that return a positive R.O.I. Until then, they keep their sales pitch to themselves.

[superfecta]

Originally posted by kgonzales
I am creating software that implements the Pizzola Hadicapping Magic Method. I use BRIS DRF mutli files for my pp info. What I am trying to do is to create a formula that will predict/create an oddsline that mirrors what the final oddsline should be based on conventional handicapping (speed, form, class, etc.). I am trying to handicap the public to know when I have true and false overlays. I'm thinking that I can just use the Prime Power numbers, but I don't really know know how I would create a properly weighted oddsline based on this number. Any ideas would be greatly appreciated.

Also does anyone know what the points per beaten length for BRIS speed numbers are at different distances. They only list approximations for sprint and route on their website.

Thanks,
Kristian:confused: Maybe the way to look at it is to establish an odds line based on your ability to pick winners.That is,review your winners and their odds at post time.Then establish that as your "favorite".When your fav and the publics fav clash,you have a legit overlay(if you pick enough winners to make a profit ITFP).Or maybe when they are comparable you have a "key" horse,(because thats as close to a lock you get).Decide then whether its worth a bet.Establish a range that will keep you in positive ROI.I don't know how effective it would be,thats for you to decide if you want to do the research.It might be helpful to get contenders at longer odds,if you play exotics.Low odds horses are obvious for the win,but which longshots have a legit chance to land in the money?Good Luck,have fun

[tanda]

Ray,

You still have not explained how you can convert power ratings to an oddsline by hand under your system but could not write code to do the same thing.

[Aussieplayer]

Hi guys,

What absolutely dumbfounds me is that whenever an oddsline thread happens, there is precious little "real" response. Which is interesting when you consider that it's (supposed) to be one of the most important areas of handicapping!

I've always been interested in the ways US handicappers make odds line - because I know the "popular" way they are done here. From what I can tell, the most popular way is to "imagine the race run 100 times then guess the percentage!" - not really heard of too much here. I'm simply curious as to whether that really is the popular method for you guys, or just one of those "book myths." I think Dick Schmidt just bets his top two as long as he can get a 50% return. No odds line at all - but at least that's honest!!
Does HSH do an oddsline?

Love being here - this curiosity of mine in this area has just got me a bit frustrated, lol

Anyway - in case of anyone's reciprocal interest, the Aussie way:

The tradition here are class/weight ratings. These result in the field being handicapped in lengths. ie. The winner, the next horse will be rated x lengths behind and so on.
From what I can tell, this is NOT the common way in the US - is that correct? Though I think Charlie Carroll does it that way with his figs.

Anyway, the top rated will be assigned 1.0
You then have a table where each possible "lengths behind" has a value. The individual horse is of course divided by the total to determine the percent chance.
Whilst I've seen different proggies use different kinds of lines (eg linear), the popular one is still Don Scott's model. How he derived these figures I'm still not sure - but at least they're weighted towards the contenders, which I'm told is important (by Pizzolla & Carroll).
Each point in Scott's rating system is worth 1/3rd of a length:
0.0 (winner) = 1.00
1 point (behind) = .9
2 points = .8
3 = .67
4 = .57
5 = .5
6 = .4
7 = .33
8 = .25
9 = .2
10 = .17

....Couple of these values might be off (don't have the book with me) but you get the idea.

If this has made for crappy, boring reading to you guys - I apologise.

Cheers
AP

[FortuneHunter]

Dear Ray,
Any method / algorithm that you can put on paper, I can incorporate into a computer program. Furthermore, if the information required to do the power rating is available in electronic form, such as a PP data file. I can make it a totally automatic operation.

Doing it by hand is a waste of time.

Computers fly jet aircraft, the Space Shuttle, play chess, control the engine in your car, runs the worlds financial institutions.

And you think your method can’t be programmed and be as accurate than if done by hand?

FH

[FortuneHunter]

On a more informative note,

A Projected Odds Line (Morning Line, Your Line, etc.) should be “balanced”. The real odds are balanced. This is because the real odds are based on money bet and the book has to be balanced or the state gets upset.

This is example of a balanced line using a point system conversion that is used by professional odds makers:
odds = Points
3/5 = 62
1-1 = 50
8/5 = 38
2-1 = 33
5/2 = 28
3-1 = 25
7/2 = 22
4-1 = 20
5-1 = 16
8-1 = 11
10-1 = 9
15-1 =6
20-1= 4
30-1 = 3

For “balanced odds line” the points should equal 100.

Horse A 8/5 (38)
Horse B 4-1 (20)
Horse C 5-1 (16)
Horse D 8-1 (11)
Horse E 15-1 (6)
Horse F 20-1 (4)
Horse G 20-1 (4)

Above is an example of a “balanced” line, because the points add up to 99.

However real Odds Makers take into account the takeout and breakage so the point total they use is 125 to 130 instead of 100.

So, if someone wants to sell you an odds line or an odds line method, be sure they balance the book. If not, it does not reflect the reality of real Mutuel Pool odds.

A full chart of the points for odds from 1/10 to 100-1 is available on request

www.handicappers-datamine.com

FH

[superfecta]

Originally posted by FortuneHunter
Dear Ray,
Any method / algorithm that you can put on paper, I can incorporate into a computer program. Furthermore, if the information required to do the power rating is available in electronic form, such as a PP data file. I can make it a totally automatic operation.

Doing it by hand is a waste of time.

Computers fly jet aircraft, the Space Shuttle, play chess, control the engine in your car, runs the worlds financial institutions.

And you think your method can’t be programmed and be as accurate than if done by hand?

FH Until some comes up with artificial intellegence that works,there is not a program that will make a novice a winner at the track.Hard for pros to do it, because at some point,decision comes into play.Although I have my doubts about Rays method,he has said the user has to decide if the horse is worth the bet.The computer can store and deliver information,but the deciding factor always comes back to the handicapper.Thats what makes it so difficult.I have seen posts saying,"well in this case this will happen",and probabilities can be figured to the enth degree,to prove it.But it doesn't always happen that way,because its just one race.
I have seen others posting saying if it wasn't for ------- I could/would win(more,more money,more races,etc.)But since you can't always change circumstances to fit your idea of what should happen or could happen,you have to play the hand you got,or quit.
Sorry FH,I didn't mean you specifically,this was just a thread that got to me and my rant..

[ranchwest]

Originally posted by FortuneHunter
Dear Ray,
Any method / algorithm that you can put on paper, I can incorporate into a computer program. Furthermore, if the information required to do the power rating is available in electronic form, such as a PP data file. I can make it a totally automatic operation.

Doing it by hand is a waste of time.

Computers fly jet aircraft, the Space Shuttle, play chess, control the engine in your car, runs the worlds financial institutions.

And you think your method can’t be programmed and be as accurate than if done by hand?

FH

Must be rocket science.

[FortuneHunter]

Originally posted by RayGordon



Way too many judgment calls are required to totally automate my method.

If it could be done, it would have been done by now.

Can you provide 1 simple example of a judgement call so I show you how easy it is to program?

In regards to Superfectas comment, I did not mean to imply that a program can take certain judgements, guess, feeling, hunch, luck out of handicapping. Or a program can make you a better handicapper.

I am trying to state that a formulas with a high degree of decision making can be programmed. And, if the formula is any good, it is worth it in time saved. No offense taken

FH

[tanda]

Well, the method could still be coded. Many exceptions could be coded as well (those for which an ascertainable standard is available). But if the method says " do this ... unless the handicapper thinks differently" and the "differently" is not an ascertainable standard, then that aspect cannot be coded. For example, if the exceptions are purely "gut instinct", then no, that part cannot coded. But if the method requires that it be applied and then "gut instinct adjustments" be made to make it work, then any system is profitable.

For example, bet any horse with the first name beginning with A but make an exception when the handicapper's judgment indicates. Well, this could work ... assuming the handicapper knew when to make an exception. In that case it is the handicapper's judgment and skill and not the system that is winning.

Ray seems to be saying "apply my method ... but make exceptions when needed and those exceptions cannot be clearly defined (that is why they cannot be coded) and if you make the right judgment calls, you will win".

Really? Who cares? If you make the right judgment calls, then you can win with any "system".

If I am selling a methodology, I should be able to delineate clear rules that show a profit ... then state that the handicapper may be able to improve those results with his judgment. But if the methodology only works if the handicapper makes completely subjective calls, then it contributes nothing since the handicapper will win with his judgment even without the system.

[GameTheory]

Because they're not programmed for that. But they could be.

Let's try the one simple example. It doens't have to be anything related to your method, just something equivalent. Let's have a contest to see who can code it best.

[GameTheory]

A man could come up with one simple example.

[ranchwest]

Originally posted by RayGordon


I've decided to keep my knowledge to myself and just cash in betting, like everyone advised me to.

This game is turning into child's play fast. I like it.

Posting here, I can see why. No one's got a clue what the new answer is. That's a good thing.

I'm not going to give it to them. Have fun in the maze.

Ray! Ray! Help! We're all floundering! Help! My computer has a stranglehold on my wallet! Help! I've fallen and I can't get up!

[FortuneHunter]

Dear Ray, what is your website address. I would like to check out your free winners (picks).
FH

[Tom]

Originally posted by GameTheory
A man could come up with one simple example.

Especially a simple man!
~G~

[kgonzales]

I hope this is not a vain attempt to get this thread back on track. I appreciate the suggestions so far, but maybe I wasn't clear in my initial query. I need help with creating a formula using traditional speed, form, class factors that will predict/handicap the public betting, not my actual value bets or contenders with a fair degree of accuracy. I could use the morning lines, but they are not consistent track to track and are always inflated so they don't accurately predict the final betting.

In his book, Pizzolla gives the example of a fairly obvious horse that rates 2-1 based on traditional/obvious factors, but shows up at say 8-1 or higher (just an example not a rule) on the board . If this horse has the top speed ratings, shows great form, and is not going up in class then there is no way he should be paying those odds unless there is something terribly wrong. To think you are so much smarter that the public in such an obvious case is suicide. If I'm going to play a 15-1 shot he better look really bad on the surface (layoff, poor form, bad connections). I want to be able to open a race and quickly (instantly) see where the the money's going to go. It would help in finding races to play and which to stay away from. My handicapping would come after deciding if a race is worth playing.

I know that ITS has the HM contention finder (or new Master Magican) which performs this function, but I am a dedicated Mac user and those only run on PC. I have seen the HMCF and old HM programs, but I don't really like their design or the numbers they generate. I like using the BRIS pace and speed numbers for creating my PBS/PPF ratings. They seem to be more consistent track to track.

So again, any suggestions would be greatly appreciated. I hope this clarifies my problem.

KG

[tanda]

Actually Ray, your involvement in this thread began with a statement that no oddsline could be calculated with a computer because it was not a binary calculation. Not that your methodology could not be coded, but that no oddsline of any type could be computed by machine. I was disputing that statement.

You have a method to convert a power rating to an oddsline. Why can that not be coded? I am assuming there is a clearly delineated process for the conversion that can be coded.

Next, you made the statement that no oddsline could be computed by machine. Can you prove that statement to be true? Hardly, since you generally cannot prove a negative.

To date, I have seen no oddsline methodology that has been verified to a reasonable degree of certainty to be accurate (say 95% confidence level). But I know of no reason why, if one exists or could exist, it is a metaphsical fact that it could not be processed by software code.

[tanda]

I reviewed the post in this thread.

It seems to me that Dave Schwartz and others weighed in with some responsive answers ... then the helpful answers abruptly stopped when Ray posted.

Coincidence?

[GameTheory]

Regarding predicting the public.

I have no automated method, and have read Pizzolla on the subject and generally agree with him. But remember, if your predictions of the public are WRONG -- that doesn't neccessarily mean you're doing something wrong.

As you just said, those are the situations you need to pay attention. The whole point of predicting the public is to know what to do when they aren't betting as they "should". Using sophisticated (presumably) numbers like the BRIS power ratings I wouldn't use because they sometimes zero in on real longshots, and only a minority of the public have them.

Just look at the obvious factors that most of the public has access to: trainer & jockey -- the top ones always get more action, Beyer figs, early speed, recent finish positions, and overall record. You should also do this sort of analysis first, so you don't cloud your mind with the more telling details that you will learn upon doing the "real" handicapping.

[Rick]

It's always interesting to me that just about every subject splits horseplayers into the two schools of handicapping as art vs handicapping as science. Laying out a set of rules in advance or programming them in software before testing a method eliminates the possibility of "improving the results" after the fact. Authors of handicapping books naturally prefer the subjective (artistic) approach and software developers naturally prefer the objective (scientific) approach.

Unfortunately, the best factors to use in creating an estimate of the public odds would change over time so their's no good answer as to what to use. Morning line odds is the only factor that incorporates the shifting public opinion and for that reason it's the best single factor. That can be verified by regression analysis, which shows that about 80% of the variance in actual odds at major tracks is explained by the morning line.

The best estimate of true odds involves using the actual odds themselves along with your own factors. William Benter of the Hong Kong Betting Syndicate proved that quite conclusively.

[raybo]

I am resurrecting a thread from 2002 on the subject of creating an oddsline that would represent the chances, for each horse in a race, of winning the race (based on a proprietary overall figure which I, obviously, have a high degree of confidence in it's accuracy).

Assume that the proprietary figure is a composite of all my handicapping factors rolled into one number.

I'm not interested in running 1000 iterations in order to achieve an oddline.

Assume a figure of 90 is, of itself, reflective of it's value over figures of 88, 83, 81, 78, 72, 63, and 55 and is reflective of it's chances of being the winner among these horses, how would I write a formula (for Excel) that would convert the figure to odds (must include the takeout % for the track involved, I can do the "lookup" portion of the formula to extract the correct takeout % for any given track).

I have seen this example but am not sure how the takeout % should be formatted, decimal or %:

((1-takeout) / (figure / total of all figures)) -1

[DanG]

I hope I don’t divert Raybo’s inquiry here (post # 36), just real quick…

One thing these old threads remind me is how much I miss “Rick” from Nevada posting. Any newer members please do yourself a favor and search his posts. Very pragmatic / intelligent approach to racing research imo. :ThmbUp:

[Tom]

Also posted under Rick Ransom. FYI.
Yes, he is missed.

[DanG]

Also posted under Rick Ransom. FYI.
Yes, he is missed.
Thanks Tom,

Time for some new searches on a rainy day.

[2low]

I am resurrecting a thread from 2002 on the subject of creating an oddsline that would represent the chances, for each horse in a race, of winning the race (based on a proprietary overall figure which I, obviously, have a high degree of confidence in it's accuracy).

Assume that the proprietary figure is a composite of all my handicapping factors rolled into one number.

I'm not interested in running 1000 iterations in order to achieve an oddline.

Assume a figure of 90 is, of itself, reflective of it's value over figures of 88, 83, 81, 78, 72, 63, and 55 and is reflective of it's chances of being the winner among these horses, how would I write a formula (for Excel) that would convert the figure to odds (must include the takeout % for the track involved, I can do the "lookup" portion of the formula to extract the correct takeout % for any given track).

I have seen this example but am not sure how the takeout % should be formatted, decimal or %:

((1-takeout) / (figure / total of all figures)) -1

Raybo,

I don't adjust my odds line for takeout. I figure since the tote board odds reflect 100% of what each horse will pay after takeout, I don't need to make any further adjustments for that. I just compare the odds I assign to a horse to the odds on the board and leave it at that.

The formula would simply be 1/((figure/total of all figures)-1)

[raybo]

Raybo,

I don't adjust my odds line for takeout. I figure since the tote board odds reflect 100% of what each horse will pay after takeout, I don't need to make any further adjustments for that. I just compare the odds I assign to a horse to the odds on the board and leave it at that.

The formula would simply be 1/((figure/total of all figures)-1)

You could be right, not using the track takeout in the formula. I was thinking that my odds should also reflect the track take to enable a better comparison against the public odds. I guess it doesn't really need to compare in that way. Simply knowing that a horse has a better or worse chance of winning than it's tote odds should be enough.

Thanks!

[cmoore]

I tried an oddsline years ago..Didn't work for me..I just look at each horses odds 2-3 minutes before post and mark down overlay or not..Then from the overlay horses I make my bet..It's a pretty sophisticated way of doing things. LOL!!!

[2low]

I tried an oddsline years ago..Didn't work for me..I just look at each horses odds 2-3 minutes before post and mark down overlay or not..Then from the overlay horses I make my bet..It's a pretty sophisticated way of doing things. LOL!!!

Hey, if you can prove to yourself that you win with your method, it's sophisticated enough!

[Overlay]

Averaging the rating numbers themselves to calculate odds would be valid if the numbers represented an accurate ratio of winning chances. Otherwise, even if higher ratings represented progressively greater winning chances, wouldn't you have to determine the actual winning percentage associated with each of the numbers, or with a horse's percentage of the total of the numbers for the entire field, and base your odds figure on that?

[Overlay]

To clarify, when I said "averaging the rating numbers", I was actually referring to adding the numbers for each horse in a race to obtain a total for the field, and then calculating each horse's odds based a comparison of its individual number to that total, minus the horse's own number (as raybo indicated, but without adjusting for the take, since toteboard odds reflect the minimum amount that a winning wager will actually pay, as 2low pointed out).

[InFront]

I tried an oddsline years ago..Didn't work for me..I just look at each horses odds 2-3 minutes before post and mark down overlay or not..Then from the overlay horses I make my bet..It's a pretty sophisticated way of doing things. LOL!!!

I also never seen an oddsline to really have any value from a betting view. In todays age you can have access to many accurate odds lines or probabilty lines whether as sold as a stand alone product or an odds line produced in one of the many popular software programs now available. I tested many of these and now use my own odds line/probabilty line which I would say is accurate as any available and have been tested on over a million horses. But this doesn't mean I can use it to make money.

Any odds line can first be proven that it is accurate by simply running it through any good size database. That is backtest 25% win rate horses, 17%, 4%, etc. and you will find if such horses win at a near perfect % as given. Great you now have an accurate odds line. Now you can simply backtest it again to prove if having such oddsline has any realvalue. Just backtest it against ANY other variable: final odds, mlo, ranked speed, ranked class, etc. You will see more often than not the original prob% goes out the window and to acheive any profits is near impossible using such a liner. We wish there was such on magical odds line where say we can find horses that win 20% of the time and they still win at that rate even if they have post odds of 7/1 or higher. Our 30% horses even win at the rate when they have 4/1 or higher odds. If only in our dreams.

[2low]

We wish there was such on magical odds line where say we can find horses that win 20% of the time and they still win at that rate even if they have post odds of 7/1 or higher. Our 30% horses even win at the rate when they have 4/1 or higher odds. If only in our dreams.

This, of course, is the key. Any good odds line must be accurate when tested against overlays it finds.

I'm testing my odds line against both the entire population of horses I consider contenders, and also separately against only the overlays. I also figure how many winners above or below my assigned percentages each odds range is over the races I've handicapped. I think underlays winning at a much higher rate than the odds I've assigned would be a sure sign of a problem with my formulas.

I don't know if I understand the rest of your post. Are you saying your 3/1 horses win 25% of the time, but not if they are higher than 3/1? If so, I'd say you do not have an accurate odds line. An accurate odds line should not vary much from one tote board odds range to another.

This does make me think I need to expand my testing though. I'm simply looking at all overlays as a group at this time. I suppose to get a real good look at how I'm doing, I need to see how my 3/1 horses do against all the odds ranges above individually. There may be a point where I have "too much" of an overlay - where I've obviously just goofed. At 20/1, my 2/1 shot is probably a mistake:D.

I should have relevant data in about July of 2027.

[2low]

On commercially available odds lines - there can't possibly be a good one out there, can there. One that hits 25% of all overlays when it says a horse is 3/1? I can't imagine selling mine if it works unless I found a tiny group willing to pony up big bucks.

Recovering lost value by setting a tool like that loose on the open market would be next to impossible I would think.

[InFront]

I don't know if I understand the rest of your post. Are you saying your 3/1 horses win 25% of the time, but not if they are higher than 3/1? If so, I'd say you do not have an accurate odds line. An accurate odds line should not vary much from one tote board odds range to another.

Any odds line can simply converted into a probability line or vice a versa. And regardless if you are looking at odds or probabilty numbers any decent "line" will win close to theie own expected rate range. Meaning your 27% win rate plays will win at about 27% of time when tested over a large sample of races. Your 2/1 shots which would also be 33% shots will win close to that rate.

Now if we agree on this up to that point when you then retest those same range rates against final post odds your win rate or probability %s will be way off. Are you saying using your own line their not???

Example backtest you 20% prob horses when they are over say 7/1 final odds, test your 27% shots when they are over 5/1 final odds, etc. You are saying that your own prob rates REGARDLESS OF FINAL ODDS still win at their original expected hit% rate over a large sample test. If so quit your job tomorrrow and start betting heavily!!!

[raybo]

Well, obviously you guys are off on some other tangent only shared by database people. I got my answer.

Thanks.

[2low]

Any odds line can simply converted into a probability line or vice a versa. And regardless if you are looking at odds or probabilty numbers any decent "line" will win close to theie own expected rate range. Meaning your 27% win rate plays will win at about 27% of time when tested over a large sample of races. Your 2/1 shots which would also be 33% shots will win close to that rate.

Now if we agree on this up to that point when you then retest those same range rates against final post odds your win rate or probability %s will be way off. Are you saying using your own line their not???

Example backtest you 20% prob horses when they are over say 7/1 final odds, test your 27% shots when they are over 5/1 final odds, etc. You are saying that your own prob rates REGARDLESS OF FINAL ODDS still win at their original expected hit% rate over a large sample test. If so quit your job tomorrrow and start betting heavily!!!

First off, reading my post again the tone seems a bit confrontational - I didn't mean it that way. I was simply stating my belief that an odds line ~should~ perform well at any tote odds. If not, I agree with you that the line cannot be used for betting purposes.

As for mine, I'm just in testing mode right now. I don't have enough race data collected to know whether it works long-term or not. So far I am making money with it, but the sample is really too small to draw any final conclusions.

[Dave Schwartz]

2low,

I was simply stating my belief that an odds line ~should~ perform well at any tote odds.

You are correct: it should. But what if it doesn't? That does not mean that it is not a valuable tool.

But I quibble...

Your posts are excellent and you obviously know (or at least believe) what you are talking about. Performance is the key.

For years I was against an odds line approach, simply because it can never really be tested. Sure, your overall performance can be tested (and one could certainly argue that is more than good enough), but you cannot look at an individual race and say, "I got this one right" any more than you could say that you got it wrong. Please note that winning a race does not prove it was a positive expectancy bet.)

Then, as my database ran into the millions of entries and I could research to my hearts content, I assume that the answer would be right in front of me at some point, so I moved towards odds lines . (We actually make probability lines, but it is, of course, the same thing.)

Now I find that I simply cannot get the accuracy that I need from top to bottom of the odds spectrum. I (and many of my users) have no problem locating profitable overlays in the medium-to-high ranges OR I can build a tool that finds occasional overlays at the low end. But getting them both together in the same line-making engine has proven to be very difficult.


Okay, with this preamble out of the way, please share whatever tips you may have to assist in this matter. I am all ears.


Regards,
Dave Schwartz

[InFront]

2low,
For years I was against an odds line approach, simply because it can never really be tested. Sure, your overall performance can be tested (and one could certainly argue that is more than good enough), but you cannot look at an individual race and say, "I got this one right" any more than you could say that you got it wrong. Please note that winning a race does not prove it was a positive expectancy bet.)

Your HSH software is one that I personally never had a chance to test and use but read much about it and seems would be as solid as others sold. But I did get a chance to test some of your other "top competitors" software which include their own probability lines. When tested over large samples they seem fairly accurate just as yours may be. But as I think you or anyone else would agree their accurate win%s go out the window once you add another variable such as post odds. For example if you test your probabilty line for say 21% horses over a large test those such horses will win close to their expected probability rate. Now if you test them at 7/1+ post odds three things will happen: far less horses would qualify, their average payoffs would go up, they don't win near that expected rate anymore. So by playing a certain prob% group only when at "overlay post odds" may not even help their final roi.

Of course if such a probability line can ever overcome that fact you would be all set for life. But what I found with all odds/probability lines is what good is it knowing that this horse has a X% chance or winning or that horse has a X% chance of winning if the final post odds on how the public bets each horse counter-effects their expected win rate. That my friend is the weakness of any odds/probability line but many players persistently buy odds/probability line reports available on the Net and some at expensive daily prices. Their better off just buying some decent software program as yours or others which would work just as good or should I say bad if they really want a odds line.

[Dave Schwartz]

Actually, our probabilities adjust nicely for the true odds.

In fact, they do so dynamically (if you so desire). That is, as the odds change, the probabilties change as well.

Our issues are as outlined above: We can win with medium-to-high odds overlays or we can win with low-odds overlays. It is difficult to do both.


WHat would your suggestions be to remedy this?

[JeremyJet]

Actually, our probabilities adjust nicely for the true odds.

In fact, they do so dynamically (if you so desire). That is, as the odds change, the probabilties change as well.

Our issues are as outlined above: We can win with medium-to-high odds overlays or we can win with low-odds overlays. It is difficult to do both.

WHat would your suggestions be to remedy this?

Your personal probability line has to be conditionally independent of of the pari mutuel line. Conditionally independent means: after I see the pari mutuel line, I don't need to alter my opinion.

JeremyJet

[2low]

Your personal probability line has to be conditionally independent of of the pari mutuel line. Conditionally independent means: after I see the pari mutuel line, I don't need to alter my opinion.

JeremyJet

I wonder about this. While this is the path I've chosen for now, the public is a pretty damn good handicapper. I could see their opinion being good data in the right system.

I also believe in "too good to be true". I can't support my claim yet, but I'm betting that if my 2/1 shot is 25/1 on the board with 0 MTP, I've probably got a problem.

[broadreach]

Odds lines for smaller tracks are sometimes a waste of time with the high fluctuations of the tiny pools.
You almost need to wait till post time to have a fair comparison.

[InFront]

Actually, our probabilities adjust nicely for the true odds.In fact, they do so dynamically (if you so desire). That is, as the odds change, the probabilties change as well.

Our issues are as outlined above: We can win with medium-to-high odds overlays or we can win with low-odds overlays. It is difficult to do both.
WHat would your suggestions be to remedy this?

So your odds line is not fixed once scratched horses are removed and adjust itself and actually uses the post odds for a final line? But why adjust them against todays post odds wouldn't you want to know how they do and accurate they are as a fixed stand alone factor?

Now when you say you can win with medium-high odds overlays do you mean the ratio of what your odds line reads vs the actual post odds? For example if you played only horses that have a odds line of 5/1 but now have post odds of 10/1+ meaning a 2-1 ratio overlay they are actually profitable longterm? Yes I understand when doing or testing this concept of course those 5/1 odds line horses will win much less than a all 5/1 test of horses but you are saying they are actually profitable???

[Dave Schwartz]

Jet,

Your personal probability line has to be conditionally independent of of the pari mutuel line.

The biggest horseplayers (and winners) on the planet would disagree with you.


Regards,
Dave Schwartz

[Dave Schwartz]

Now when you say you can win with medium-high odds overlays do you mean the ratio of what your odds line reads vs the actual post odds? For example if you played only horses that have a odds line of 5/1 but now have post odds of 10/1+ meaning a 2-1 ratio overlay they are actually profitable longterm? Yes I understand when doing or testing this concept of course those 5/1 odds line horses will win much less than a all 5/1 test of horses but you are saying they are actually profitable???

I am saying that horses which project a $net above $2.00 are returning a $net above $2.00 and are, therefore, profitable.

The problem is that when I design a system that is accurate above (say) 5/1 it does a terrible job with horses under 5/1. That is, the low-priced horses that say "profit" are losers.

If, instead, I build a system which produces profit in the lower ranges, say under 4/1, it is no good at producing profit in the higher ranges.


Regards,
Dave Schwartz

[InFront]

I am saying that horses which project a $net above $2.00 are returning a $net above $2.00 and are, therefore, profitable.

The problem is that when I design a system that is accurate above (say) 5/1 it does a terrible job with horses under 5/1. That is, the low-priced horses that say "profit" are losers.

If, instead, I build a system which produces profit in the lower ranges, say under 4/1, it is no good at producing profit in the higher ranges.

That is good you can even find things that are longterm profitable in low odds ranges. What I have found is overall I find it very difficult to produce a method that is profitable when mainly dealing with low odds horses. Doesn't matter if it is some odds line, complicated algorithm, simple formulas, etc. It seems the public is so "in tune" when it comes to low odds horses and that they never outrun or win at a higher rate than how the public bets them. While on the otherhand if mistakes are gonna be made by the public it is usually among higher odds stuff, no doesn't have to be huge longshots but even middle priced horses.

This can be simply proven by taking any decent method that is "overall profitable" and then query it based on final odds groups. Most of the time you will notice most if not all of the profit came from betting those higher odds horses. I would love to find something that is profitable that works with low odds stuff cause I know such a method will also have a nice win%. But as you know when dealing with higher odds methods even if we can sweak out a profit we have to suffer those terrible low win%s along with long losing streaks.

[Dave Schwartz]

This can be simply proven by taking any decent method that is "overall profitable" and then query it based on final odds groups. Most of the time you will notice most if not all of the profit came from betting those higher odds horses. I would love to find something that is profitable that works with low odds stuff cause I know such a method will also have a nice win%. But as you know when dealing with higher odds methods even if we can sweak out a profit we have to suffer those terrible low win%s along with long losing streaks.

Yes. I have found the same thing. I can play a significant number of low-priced horses and be profitable. However, as a rebate player, it is very difficult to get the necessary handle if you skip races.

If playing multiple approaches was an easy alternative I would move towards that. However, that would entail a great deal of re-write.


The logic, of course, bothers me: As stated earleir, an odds line should work equally well througth all odds ranges. If I could make that happen, then the world would become my oyster.


Dave Schwartz

[rrbauer]

I'm thinking that I can just use the Prime Power numbers, but I don't really know know how I would create a properly weighted oddsline based on this number. Any ideas would be greatly appreciated.



Many, many flaws with Prime Power #'s and Bris knows it (because I've been in their ear about it!). Mixing sprints/routes/turf/dirt/poly to develop the number is wrong. The number needs to reflect what today's scenario is and be developed accordingly. Also, in that same vein, a par PP# for today's race would be very useful to give you some idea of "field strength". I question PP's value as the sole stat for developing an odds line.

Most of the "computer" handicappers build a probability estimate for each horse and then normalize that to 1 (or 1.25 to compensate for takeout) and then do a prorata allocation of those points to each horse and convert the resulting percentages to odds/1. That is their baseline for identifying overlays and betting. You probably know all of that anyway.

With all of the computer data floating around, I would try for openers to develop a correlation between track morning lines and final odds. This will help you filter out the tracks that do a decent job with ML odds versus the ones that are just apecrap. If the track(s) that you play have a good ML maker then that's what I would use for my baseline odds. It's a vexing problem particularly for tracks with small pools where a $100 bet will create havoc with the odds.

I didn't read all the posts on this thread, so if somebody has beat me to an idea, consider this "confirmation". :)

[Overlay]

I think that it helps if you can come up with a plausible explanation as to why a particular horse is being overlooked in the betting. If it seems clear to me that a horse that I like is a relatively high overlay compared to my line because of the public giving too much weight to a positive factor possessed by another horse in the race, and overbetting it as a result, I don't let my choice's odds deter me from backing it.

[Overlay]

Now I find that I simply cannot get the accuracy that I need from top to bottom of the odds spectrum. I (and many of my users) have no problem locating profitable overlays in the medium-to-high ranges OR I can build a tool that finds occasional overlays at the low end. But getting them both together in the same line-making engine has proven to be very difficult.

Quirin mentioned the same problem with low-odds horses in presenting the odds model that accompanied his multiple regression formulas in Winning at the Races. He commented, "Multiple regression formulas tend to be conservative. Better than 50% of all regression selections [that is, top-ranked horses] will be underlays. And very few horses will be rated as 'worth a bet' when their toteboard odds are below 2-1. This is the computer's way of telling us that very few horses are worth low odds." (However, he noted at the same time that the two specific regression formulas for sprints and routes that he discussed in the book proved to be equally profitable (at that time), whether play was restricted to top-rated horses with toteboard odds above 2-1 or not.) Of course, much has changed since then (smaller field sizes, available handicapping data, and so on), mostly with the effect of further reducing the toteboard odds of horses on the low end of the spectrum.

[TrifectaMike]

Yes. I have found the same thing. I can play a significant number of low-priced horses and be profitable. However, as a rebate player, it is very difficult to get the necessary handle if you skip races.

If playing multiple approaches was an easy alternative I would move towards that. However, that would entail a great deal of re-write.


The logic, of course, bothers me: As stated earleir, an odds line should work equally well througth all odds ranges. If I could make that happen, then the world would become my oyster.


Dave Schwartz

Getting an oddsline to work equally well through all odds ranges is extremely difficult. The problem is that the models suffer from non-linearities. The differentials of the factors are not linear for all odds ranges. This has the effect of introducing large amounts of noise. Often more noise than real data, which reults in poor estimates.

Mike

[raybo]

Quirin mentioned the same problem with low-odds horses in presenting the odds model that accompanied his multiple regression formulas in Winning at the Races. He commented, "Multiple regression formulas tend to be conservative. Better than 50% of all regression selections [that is, top-ranked horses] will be underlays. And very few horses will be rated as 'worth a bet' when their toteboard odds are below 2-1. This is the computer's way of telling us that very few horses are worth low odds." (However, he noted at the same time that the two specific regression formulas for sprints and routes that he discussed in the book proved to be equally profitable (at that time), whether play was restricted to top-rated horses with toteboard odds above 2-1 or not.) Of course, much has changed since then (smaller field sizes, available handicapping data, and so on), mostly with the effect of further reducing the toteboard odds of horses on the low end of the spectrum.

Excuse my ignorance, but what the hell is multiple regression (as a matter of fact, what the hell is a regression), as it pertains to creating a basic oddsline based on one composite rating? Try to speak layman's English please. Oh, and assume no knowledge of statistics or probabilities.

[TrifectaMike]

Excuse my ignorance, but what the hell is multiple regression (as a matter of fact, what the hell is a regression), as it pertains to creating a basic oddsline based on one composite rating? Try to speak layman's English please. Oh, and assume no knowledge of statistics or probabilities.

In simple terms, regression is simply fitting a line or curve to a bunch of data points subject to some error criteria. For example, if you were to plot speed rating vs finish position, and you wanted to find the relationship between speed and finish postion. An approach would be to use regression analysis subject to a least squares fit criteria. Basically making some sense from a bunch of data points.

Mike

[raybo]

In simple terms, regression is simply fitting a line or curve to a bunch of data points subject to some error criteria. For example, if you were to plot speed rating vs finish position, and you wanted to find the relationship between speed and finish postion. An approach would be to use regression analysis subject to a least squares fit criteria. Basically making some sense from a bunch of data points.

Mike

Wouldn't it be simpler to just say "graphing" data points, without the graphics?

Most of us had to learn to graph at least by juinior high school.

[Overlay]

The main point that I was attempting to make was a confirmation of previous observations about the difficulty of creating an odds line that consistently locates overlaid horses in low odds ranges. The fact that the particular handicapping approach in question was a multiple regression formula was secondary to that point. However, the wording that I used was not my own. It was that of the reference I cited, which was (and, I would think, still is) regarded as a major contribution to handicapping literature, however dated some of its findings may now be. Pardon me if I made unwarranted assumptions concerning its use, or its familiarity among board members.

[dutchboy]

Excuse my ignorance, but what the hell is multiple regression (as a matter of fact, what the hell is a regression), as it pertains to creating a basic oddsline based on one composite rating? Try to speak layman's English please. Oh, and assume no knowledge of statistics or probabilities.

http://www.statsoft.com/textbook/stmulreg.html

[raybo]

Averaging the rating numbers themselves to calculate odds would be valid if the numbers represented an accurate ratio of winning chances. Otherwise, even if higher ratings represented progressively greater winning chances, wouldn't you have to determine the actual winning percentage associated with each of the numbers, or with a horse's percentage of the total of the numbers for the entire field, and base your odds figure on that?

Just "got" this one. Took me a while to figure out what you were saying. If the ratings that the oddsline (win probability line) is based on range from 0 to 100, then wouldn't that be an accurate ratio of winning chances?

Even if they weren't 0 to 100, as long as you have a lowest rating and a highest rating possible then there should be a formula that would convert the set of ratings into a probability/win odds line.

There is, at least in the case of my ratings, a direct relationship between one rating number and another, and it is a linear relationship.

[Overlay]

Just "got" this one. Took me a while to figure out what you were saying. If the ratings that the oddsline (win probability line) is based on range from 0 to 100, then wouldn't that be an accurate ratio of winning chances?

Even if they weren't 0 to 100, as long as you have a lowest rating and a highest rating possible then there should be a formula that would convert the set of ratings into a probability/win odds line.

There is, at least in the case of my ratings, a direct relationship between one rating number and another, and it is a linear relationship.

By referring to a ratio, I was thinking of a case where a horse with a rating of 100 would have twice as great a chance of winning as a horse with a rating of 50, and four times as great a chance as a horse with a rating of 25 (which I take you to be saying applies to your ratings). That would make it valid to base the odds line on each horse's proportion of the total of ratings for the field. Otherwise, if such a relationship weren't present, you'd first have to calculate the winning percentage associated with each individual rating, and then base your odds line on the ratio of those percentages, rather than on the ratings themselves.

[raybo]

By referring to a ratio, I was thinking of a case where a horse with a rating of 100 would have twice as great a chance of winning as a horse with a rating of 50, and four times as great a chance as a horse with a rating of 25 (which I take you to be saying applies to your ratings). That would make it valid to base the odds line on each horse's proportion of the total of ratings for the field. Otherwise, if such a relationship weren't present, you'd first have to calculate the winning percentage associated with each individual rating, and then base your odds line on the ratio of those percentages, rather than on the ratings themselves.

Ok, I think I understand what you 2 were talking about now, concerning the low odds % vs the higher odds %.

[DanG]

what the hell is multiple regression (as a matter of fact, what the hell is a regression).
Game Theory gave a great explanation (the guy really should teach) a while back. I believe this is complete from a copy. It’s the first time I’ve heard it explained where it made any real sense to my limited public education.

This might not be specifically what you’re asking, but for those who haven’t read it / the guy has one of the best minds on the board imo.

Game Theory ~ Understanding the basics is fairly easy if you can understand a simple linear regression equation:

y = a + b*x

where a is a constant and b is the slope of the line. For instance, a simple equation like this can convert Celsius temperatures to the Fahrenheit scale:

F = 32 + 1.8*C

We've got a constant (32) and are giving a "weight" (coefficient) to the celcius factor of 1.8.

We all did this stuff in school, I imagine.


Well, multiple linear regression does the same thing, except with many factors instead of just one (each factor is multiplied by a weight and everything is summed together along with the constant), and usually with problems that don't have exact correspondence like the two temperature scales do. So your answers become predictions instead of conversions. The problem with linear regression is that it is not bounded -- it is a straight line that can go on forever. So the next step is to go non-linear, which is what logistical regression does -- now your final answer is on a s-shaped curve that is bounded between 0 and 1, which happens to be a good thing when you want to predict probabilities.

But you're still using a constant and a then a coefficient/weight for each factor. It just adds some some extra mathematical magic to put it on a log scale. So it is pretty easy to grasp the basic idea conceptually. If you want to understand how to actually arrive at and calculate the weights (besides just plugging it in to your statistical software, which is what you do in practice), then you've got to read those dummies books on statistics and probability to get some background.

The Bayesian stuff is even easier to understand in concept. You simply take the conditional probability of each factor and multiply them all together. What's a conditional probability of a factor? Well, it is the probability of seeing one thing given that we've already got another thing. (Bayesian factors are simply "yes" or "no" -- we've got it or not, so if you're dealing with a factor like speed figures, you've got to threshold it or break it up into bins -- e.g. is the figure above 80? yes or no).

So we're interested in picking winners, so that's our "given" thing -- that the horse won. So like when we do impact values, we ask the question, "Among winners, what percentage of them had X factor?" The answer is the conditional probability of that factor, e.g. 50% of winners were in-the-money last race. So the conditional probability of the factor "being in-the-money last race" is 50%. Take a bunch of such factors, multiply them together, and you'll get a number for each horse that you can compare to the other horses in a race. In this extremely simplified explanation, we won't worry about the fact that final numbers won't be normalized (add up to 1.0) or worry about which factors to choose, etc. In practice, the math would be slightly more complicated, but not much.

Now, Bayesian inference assumes that your factors are all independent (don't measure the same thing in any way), which they never are, but it works surprisingly well anyway for general applications -- it is used mainly in medical research. [ If more actual doctors understood it, there would be a lot less misdiagnoses because unfortunately most doctors (much less patients) don't understand that if you take a medical test for some disease that is 90% percent accurate and it comes up positive it does NOT mean you have a 90% percent chance of having that disease. It means that *if* you have the disease, there is 90% chance of the test coming up positive. But when you consider the fact that most people don't have the disease along with the "false positive" rate of the test, it usually means your chance of having the disease is still quite small if all you're going by is the test. ]

The problem with using Bayesian inference in its simplest form as given here (called "naive Bayes") is the yes/no 1/0 nature of the factors, and of course choosing the factors themselves. It is probably more appropriate to use as a benchmark than for use in practice where other methods will likely work better.


The problem with any of these statistical handicapping methods is that it is hard to narrow in on "where the profit is". They do great at general handicapping, but you've got to beat the crowd, not just pick reasonable probabilities...

[raybo]

Game Theory gave a great explanation (the guy really should teach) a while back. I believe this is complete from a copy. It’s the first time I’ve heard it explained where it made any real sense to my limited public education.

This might not be specifically what you’re asking, but for those who haven’t read it / the guy has one of the best minds on the board imo.

I agree, having read and discussed many of his posts, he is, indeed, one smart dude. If I were going to bet on anyone posting here being successful as a handicapper, he'd definitely be one I would bet on.

[Overlay]

We wish there was such on magical odds line where say we can find horses that win 20% of the time and they still win at that rate even if they have post odds of 7/1 or higher. Or 30% horses even win at the rate when they have 4/1 or higher odds. If only in our dreams.

I understand your point completely, and the "wisdom of the crowd" certainly can't be overlooked. But in thinking about this issue, I keep coming back to handicapping factors (such as some of those that Quirin found in Winning at the Races) that have a winning percentage that is so much higher than the expected rate (based on the odds of the horses involved) that their performance can't be accounted for by random variation, and that are thus independent variables that aren't affected by interaction with other handicapping elements, and that don't rely on that interaction for their effectiveness. If there were any prospect of constructing an odds-insensitive betting line, might it not be based on the proper mix of such select factors?

[HUSKER55]

That being the case I have a question for you math guys. Suppose you know your odds on winners comes at (<9:1). Anything above that happens, say 6 times out of a hundred. (just for arguements sake)

If your odds line zero in on that number wouldn't the deviation from the strainght line be considerable less. (less shotgun effect).

Then wouldn't your low odds horse be more closely in tune with the public odds?

I am thinking that any odds line has to be in tune for you and there is no "one true odds line" until after the race.

Am I lost,..again.

husker55

[gm10]

First of all, most of you know this, but an oddsline reflects the wininng chance of each horse in the race (for example, ignoring take out, 3/1 = 25% chance of winning). The true winning probability is an unknown quantity, and we can only estimate it.

The morning-line guy makes such an estimate. The betting market makes a similar guess. To make a profit, you MUST outperform the betting public's estimate in the set of races where you decide to have a bet. To outperform means that you are better at estimating how likely a horse is to win - it does not mean you must find more winners. Merely outperforming the betting public is necessary but not sufficient, you have to overcome the 18% takeout.

This is why an oddsline is essential.

Secondly, how do you make one on the basis of raw numbers, such as speed ratings? There are a few ways to do it, but if you want a nice mix between simplicity and sophistication, I would suggest the multinomial logit model. I use this (see my website), and I find it more than adequate. In the end, the formula to calculate probabilities is fairly straightforward to program, even in Excel.

[gm10]

Jet,



The biggest horseplayers (and winners) on the planet would disagree with you.


Regards,
Dave Schwartz

Sadly, they would be mistaken. Of course, you get higher accuracy by adding live odds or morning line odds as factors to your oddsline-generating model ... but your bottom line tumbles. You are diluting your personal insights to a degree where they can no longer be profitable.

[sjk]

Sadly, they would be mistaken. Of course, you get higher accuracy by adding live odds or morning line odds as factors to your oddsline-generating model ... but your bottom line tumbles. You are diluting your personal insights to a degree where they can no longer be profitable.


I am not sure what factual basis you have for saying Dave and his high-betting friends are mistaken. My experience agrees with his.

In making a bet/no bet decision I require that the toteboard offer an 80% premium to my fair price calculation. This 80% figure is not pulled out of the air. I backtested a hundred thousand or so races to see what number gives me maximun profit and that is the answer that is right for my process.

Clearly using such a factor amounts to incorporating tote odds in the decision process and manipulating the factor up and down is logically equivalent to using the public opinion to a greater or lesser extent.

[Overlay]

In making a bet/no bet decision I require that the toteboard offer an 80% premium to my fair price calculation. This 80% figure is not pulled out of the air. I backtested a hundred thousand or so races to see what number gives me maximun profit and that is the answer that is right for my process.

If (as I've heard in some anecdotal estimates) even the best odds line by any individual can account for only 55% of the total available information about a horse's winning chances, then (assuming as a worst case that all of the remaining 45% of unknown information gravitates against your line on a particular horse), I figure that the amount of odds premium that you'd have to require to compensate for that would be 45/55, or 81.8181...%, which is very close to the 80% figure that your results indicate.

[gm10]

I am not sure what factual basis you have for saying Dave and his high-betting friends are mistaken. My experience agrees with his.

That's a fair comment.

Consider this


a) I have fitted my model with all my personal factors (100+) WHILE IGNORING ALL ODDS, BE IT LIVE TOTE OR MORNING LINE ODDS (races where some horses have no PP's are ignored here).
b) I have fitted my model with all my personal factors AND the morning line odds as an extra factor.
c) I have fitted my model with all my personal factors AND the final tote odds as an extra factor.
d) I have fitted a similar model wit ONLY the final odds as an input variable.

Assume that there is a measure of preditictive power called R2.
Then:

a) performs nearly as good as d) ... this is not straightforward, it ignores ALL betting information. So not only does the model need to interpret public data as well as thousands of people combined, it also need to overcome the lack of private information which is held by some of those bettors.
b) outperforms a) and d)
c) outperforms a) b) and d)

Tremendous. c) is clearly the way to go in terms of predicitve power. Followed by b) d) and a).

But wait. Does that correspond to a better bottom line for each scenario? No it doesn't, and it shouldn't come as a surprise either. c) gives me really awesome strike rates, but it only makes me break even. So my personal factors make up for the take-out, but not more. It's similar for b). For d) the bottom line is obviously -16-17%. For scenario a), the least predictive setup, but the most personal one, the bottom line is .... +10% ROI per dollar bet. In other words, my model may pick less winners but it picks winners that the crowd is not picking, leading to bigger pay-offs.

It is technically possible that someone has personal handicapping factors that actually create higher payoffs when combined with public odds, but in 5 years of playing with different combinations of personal factors, I have never seen this. In fact, I doubt whether the people who claim that this is true for them, have actually formally tested this. I'm open to any sort of discussion or feedback on this. Maybe I've been doing something wrong all this time.



In making a bet/no bet decision I require that the toteboard offer an 80% premium to my fair price calculation. This 80% figure is not pulled out of the air. I backtested a hundred thousand or so races to see what number gives me maximun profit and that is the answer that is right for my process.



In this sense I agree with you. This is making use of the public betting information. My personal required premium over the final odds is the take-out rate. It is sort of reassuring that this works, because it suggests that my personal probabilities are good estimates of the true probabilities.

I have tried setting it to 50%, trying to find fewer but more profitable bets, and while there are of course far fewer, there tend to be quite a few whacky ones which are clearly only there because of some mistake or past perfromance fluke. Apart from having a higher rate of stupid selections, setting the required premium to 50% simply produces fewer bets, which means that even if the profit margin is higher, the total profit may not be.


Clearly using such a factor amounts to incorporating tote odds in the decision process and manipulating the factor up and down is logically equivalent to using the public opinion to a greater or lesser extent.

Clearly it does.
However, the gentleman's statement was that the big players would disagree with "Your personal probability line has to be conditionally independent of of the pari mutuel line."
Personally I support this claim, and I have explained why.

[Dave Schwartz]

GM,

Imagine that your odds approach handicaps two horses, both with a 30% chance of winning. One horse is even-money and the other is 30/1. Only in some kind of dreamworld are those two horses equal.

The reality is that the 30/1 shot needs his probability reduced and the 1/1 horse needs his increased.

Now, I am not saying that the 30/1 horse is not a good bet. Neither am I saying that the 1/1 horse is not a bad bet.

I am simply saying that horses "like this one" at 30/1 will not hit at the same clip as the 1/1 horse in the long run.

That is why "tote smoothing" (as we have come to call it) is a good approach. Bill Benter used it in Hong Kong. We use it in HSH (both a version of Benter's approach and our own).


Regards,
Dave Schwartz

[TrifectaMike]

GM,

Imagine that your odds approach handicaps two horses, both with a 30% chance of winning. One horse is even-money and the other is 30/1. Only in some kind of dreamworld are those two horses equal.

The reality is that the 30/1 shot needs his probability reduced and the 1/1 horse needs his increased.

Now, I am not saying that the 30/1 horse is not a good bet. Neither am I saying that the 1/1 horse is not a bad bet.

I am simply saying that horses "like this one" at 30/1 will not hit at the same clip as the 1/1 horse in the long run.

That is why "tote smoothing" (as we have come to call it) is a good approach. Bill Benter used it in Hong Kong. We use it in HSH (both a version of Benter's approach and our own).


Regards,
Dave Schwartz

Tote smoothing may have worked for Benter in Hong Kong. I don't question if it did or did not. However, the purpose of an oddsline is to estimate as closely as possible within an acceptable error the objective probabilities.

The subjective probabilities as determined by the tote reflects an aggregate of handicapping factors, not unlike those used in generating your oddsline.

Therefore, by using the tote odds you are introducing unacceptable correlation in the use of your personal factors.

Back to Benter and Hong Kong. Sure it might have worked, because the tote odds were likely not reflective of his model factors.

Mike

[JeremyJet]

That is why "tote smoothing" (as we have come to call it) is a good approach. Bill Benter used it in Hong Kong. We use it in HSH (both a version of Benter's approach and our own).

Regards,
Dave Schwartz

The person I learned from was offered a position by Benter.

Btw, "conditionally independent of of the pari-mutuel line" doesn't mean no adjustments are made to the original line. Adjustments are made based on historical data.

What is "tote smoothing" based on? How do you know how much to "smooth" the line if it's not based on any historical performance data?

Regards,

JeremyJet

[raybo]

Back to Benter and Hong Kong. Sure it might have worked, because the tote odds were likely not reflective of his model factors.

Mike

Or possibly it might have worked because the pools are huge there and he was looking for many, many plays per day and striving for a consistent ROI even though very low. His volume enabled him to turn that very small ROI into huge profits. He did use his own proprietary model factors, as has been verified, but then don't most of us here? Proprietary model factors aren't the "be all and end all" to being successful at this endeavor.

ps: I'm not being confrontational here, just proposing a possible alternative to your statement.

[Robert Fischer]

Tote "smoothing" seems like a good idea to me in general.

Some of it depends on how well you understand the public's(Tote's) reasoning for that particular race.

when i wouldn't smooth:
There is a big difference between a horse that you have at 7-2 being 9-1 for no good reason, and a horse that is 9-1 BECAUSE OF: (another horse's fan appeal, rider fell at the gate last time, etc...)
In other words you have a strong opinion that he is a 7-2 shot, and you know before hand that he will probably be 9-1 at post. Not one of those situations where you say "why is he ICE COLD on the tote??"

If you don't happen to know exactly why, it is a good time to smooth.

[Dave Schwartz]

The general idea is simply to "smooth" the outliers; the extremes.

In my approach, which differs significantly from Benter's, one simply designates a minimum and maximum $net for each odds range. If a horse comes up too high or too low, then the probability is changed to match the $net.

Example:

Suppose you have a horse that is 1/2 and you have given him a 10% chance of winning. There probably hasn't been a 0.50:1 horse that bad since the beginning of time. So...

$Net = Payoff x HitRate = $3.00 x 10% = $0.30

Suppose you have designated horses in the 1/2 range as having a minimum $net of $1.20. Therefore...

HitRate = $MaxNet / Payoff = $1.20 / $3.00 = 40%

So, your 10% figure is adjusted up to 40%. The horse is still a bad bet but at least it did not make every other horse in the race a good bet.


Conversely, suppose you have a 30/1 with a 30% hit rate...

$Net = Payoff x HitRate = $62 x 30% = $18.60

... and you have set the maximum $net at $8.00.

HitRate = Max$Net / Payoff = $8.00 / $62.00 = 12.9%


Still a very profitable play but probably more in line with reality.

The final step is to re-normalize the field to 100%.

[gm10]

GM,

Imagine that your odds approach handicaps two horses, both with a 30% chance of winning. One horse is even-money and the other is 30/1. Only in some kind of dreamworld are those two horses equal.

The reality is that the 30/1 shot needs his probability reduced and the 1/1 horse needs his increased.

Now, I am not saying that the 30/1 horse is not a good bet. Neither am I saying that the 1/1 horse is not a bad bet.

I am simply saying that horses "like this one" at 30/1 will not hit at the same clip as the 1/1 horse in the long run.

That is why "tote smoothing" (as we have come to call it) is a good approach. Bill Benter used it in Hong Kong. We use it in HSH (both a version of Benter's approach and our own).


Regards,
Dave Schwartz

Hi Dave, as I said I have tried this, and the bottom line was always negatively impacted. I would pick more winners, though.

[jasperson]

I am creating software that implements the Pizzola Hadicapping Magic Method. I use BRIS DRF mutli files for my pp info. What I am trying to do is to create a formula that will predict/create an oddsline that mirrors what the final oddsline should be based on conventional handicapping (speed, form, class, etc.). I am trying to handicap the public to know when I have true and false overlays. I'm thinking that I can just use the Prime Power numbers, but I don't really know know how I would create a properly weighted oddsline based on this number. Any ideas would be greatly appreciated.

Also does anyone know what the points per beaten length for BRIS speed numbers are at different distances. They only list approximations for sprint and route on their website.

Thanks,
Kristian:confused:
Here is some of my data on bris pp. The roi is the average return for a $2.00 bet.
all 6529 races including turf races
IV roi win%
1st 2.39 $1.60 30.2
2nd 1.64 $1.76 19.7
3rd 1.21 $1.68 14.9
4th .80 $1.54 13.7
5th .42 $1.33 19.4

4th is all horses that finished in the front half of the field that were not 1st,2nd, or 3rd. 5th were all horses that finished in last half of the field.
Also I ran the same numbers for dirt sprints and routes and the number were about the same at least they were close enough for government work:D
About the only use I can see for this data is if a horse finish in last half of the field in his last start don't use him
Jack