Quantifying tepidity

[GameTheory]

How's that for a subject title?

Anyway, I'm playing with my toteboard analysis program, and I want some opinions on what would be a good way to give a "contentiousness" score to a race, using only the information available on the tote from the opening flash (20 minutes to post) to 2 mins to post, when a bet decision needs to be made. (Actually, the decision is made at 1 minute to post, using the data from 2 mins and up.)

So the data available is obviously: the odds, the pools, the percentages, the ratios of pools, etc. -- the raw form of everything is of course the actual number of dollars in each pool. Add to that the morning line, the size of the field, and the changes that occur over time; but let's not concern ourselves with any "real" handicapping for the moment.

Now, if you're like me, when you see the favorite at 3-1 or higher at 3 minutes to post, you think to yourself, "Ooooh, there's money to be made on this race." Even if you haven't handicapped the race, you can feel a big payout coming. This is what I call a "tepid favorite" (or lukewarm favorite) race, and it happens when the public doesn't know what to make of the race. Usually there's a bunch of horses around 3-1 to 5-1. Now, normally in the last couple of minutes, they decide to go with the two lowest morning line horses and their odds go down, but these are generally terrible bets. I haven't researched it, but I bet you could box the 4-5-6 odds horses in the exacta in these races and make a profit.

So it would be a useful thing to flag races where there is likely to be a medium to long shot payout. I was thinking I would just use the odds of the favorite (maybe averaged over time) as my score, then I could experiment at what level of tepidity a possible play become favorable, or whatever. But right now I'm just concerned with coming up with a way to make a number.

My simple method is probably good, but I'm wondering if any one has given this issue any thought, and come up with something more sophisticated? (Maybe using the field size somehow? I would consider a 3-1 favorite in a 12-horse field much more well-regarded than a 3-1 favorite in a 6-horse field. And of course there's no need to restrict ourselves to looking at the favorite.) Since the computer is taking care of it all, complexity of calculations right before post time is not an issue, except for the fact that simple, elegant solutions usually work better in most cases...

[GameTheory]

I knew that title was going to scare people off...

[stuball]

:confused:
Hi GT I would think that if we could have a program to chart the odds as they progress we could isolate weak favorites. We could save the odds charts and sort races by number of entrants.
then we could see what a strong fav is as opposed to a weak one....I have also thought that we could isolate betting stables to see what their betting pattern is by watching odds changes and results.
Let me know what you think of my ideas -- sounds
like we have some of the same ideas...also someday maybe you would like to think of a online betting program based on preset parameters but able to be manually controlled...Just another thought???

Looking forward to the parser..and the future.

Have a nice day !!!!!!!

Stuball

[hurrikane]

Interesting idea. I suppose you would want to include how far the odds going off have moved from the morning line. Somehow you would have to quantify the ability of the mlo maker. Some are very strong and some are pigs.
Maybe a comparison of the chaulk to the next 3 and the spread between them.
The only problem I see is that the only way to make a valuable test is to run it over a few thousand races and check your results. How you go about including all the variables in a reasonable mannor is the puzzle.
Even accumulating the data is a rather large job.

[BillW]

GT,

How about designing some research software to collect data for you. Then an IV/ROI analysis can provide some hard data . Another possible fallout of a project such as this is that you may be able to identify a "signature" attributable to each track that uniquely identifies the characteristics of the bettors at that track. (my guess (swag) is that the tote characteristics vary from track to track, but may be fairly constant at a particular track).

Good luck,

Bill W.

[stuball]

My suggestion for data accumulation is to form a partnership of some sort with each person responsible for certain tracks....I have often thought of forming some sort of club or partnership if we can get the programs to gather the data and some dedicated and reliable partners to be responsible for their portion of the data to be shared with all partners..if we could set up such a club we could have a very powerful tool. The problem with this being that it is hard to find enough dedicated people with the time to pull this off...THINK ABOUT IT. 1 persons data X number of people gathering data X the power of numbers === unable to calculate that high.
This has always been a latent dream of mine but we could make this happen as there are a lot of smart and dedicated people on this board.

OOPS THERE I GO DAYDREAMING AGAIN !!!!

Have a nice day.....

Stuball

[BillW]

Originally posted by stuball
My suggestion for data accumulation is to form a partnership of some sort with each person responsible for certain tracks....I have often thought of forming some sort of club or partnership if we can get the programs to gather the data and some dedicated and reliable partners to be responsible for their portion of the data to be shared with all partners..if we could set up such a club we could have a very powerful tool. The problem with this being that it is hard to find enough dedicated people with the time to pull this off...THINK ABOUT IT. 1 persons data X number of people gathering data X the power of numbers === unable to calculate that high.
This has always been a latent dream of mine but we could make this happen as there are a lot of smart and dedicated people on this board.

OOPS THERE I GO DAYDREAMING AGAIN !!!!

Have a nice day.....

Stuball

... Continuing the dream

I wonder how easy it would be to do a seti@home type thing. Rather than data processing in the background, this would be data gathering in the background. Rather than a lot of dedicated people, all that would be needed is a lot of people to leave their computers on line while their track was running ... minimize the need for dedication to 3 minutes a day.

Bill

[GameTheory]

Well,

I've already made a program to grab all the toteboard data on each flash and save it. (It saves all the tracks they have on racingchannel.com, which is pretty much all of them.) I've been saving all the win-place-show pool info since Feb. 6. (Haven't gotten around to adding the exotics yet, which are possibly more important, as that's where the money is going these days.) And I even made it so it I can chart the odds on a graph so I can see the patterns "visually" (not real time yet, but after the fact). There are some thoughts on the subject at:

http://www.camptownraces.com
(Look under "Practice Races")

But they aren't that detailed since he's got books to sell about it. I have certainly noticed some potentially profitable patterns, but they seem a bit shaky. I get the feeling that I need to put each pattern in context. Instead of saying, "Look, pattern X on horse A: make a bet!", I need to look at the whole race, and say, "Ok, we've got a vulnerable favorite, *now* we can look for pattern X."

But I guess the consensus is just study the data I've got, and figure it out. I was just looking for a good starting point...

[stuball]

I love when we get on this subject as I would be very glad to be a part of such a project...anybody else up for something like this? We would have to protect the data so it could not be sold or distributed in some way which I see as a potential problem. Any one game for something like this speak up....

Have a nice day !!!!!!

Stuball

[PaceAdvantage]

This has been an area of interest of mine as well, but I've never done anything concrete as far as research or program creation.

I remember a program came out recently called ToteBoard daytrader or something. Anyone ever check that program out?? It looked to be along the lines of what is being talked about here....



==PA

[GameTheory]

The website for that one is:

http://www.toteboard-daytrader.com/

It looks like a decent program, but I preferred to make my own since I am able, and can put in whatever formulas I desire. It also seems to focus on the "inefficient" markets approach, where I'm more interested in predictive patterns. Plus all my data is saved to a database for independent analysis by the user -- I don't know if the Daytrader program does that or not. Does anybody use it?

I was thinking of offering mine as a commercial program (I can't make everything for free) under a similar arrangement -- probably 10 bucks a month. That is a ways off though -- mine is pretty crude at the moment....

[gumbo]

I have made a small study of Charles Ubil's "pattern recognition" approach to tote board analysis that GameTheory mentioned earlier in this thread at my "home" track (Tampa Bay Downs)

Approximately 300 races.

Those who have been to the site know that he has presented 10 so-called "pattern races" that one can identify by tote-board analysis.

My experience is that in approximately 220 of these races, there are multiple patterns that can be identified based on his "analysis". In some cases the tote board can represent as many as 4 of his patterns. Each of these patterns have a different betting strategy. As GT stated, he's got to keep some of it to himself in order to sell books.

There may be a silver lining, however. I have found some trends that seem to support comparison of "late" betting moves (between 6 and 2 MTP) relative to M/L odds. Much more work to be done, but I would also be willing to commit time to some of the monitoring strategys presented in this thread. The "SETI" concept, in particular, has tremendous possibilities. Let's put our computers to work instead of allowing them to have all of that idle time.

Don

[GameTheory]

A few things:

First, and specifically about Tampa Bay -- there were about 5 nice priced winners on Don Rice horses recently, all with early money, and most ridden by the same jockey (starts with a P, can't remember -- Walker's the other one). Probably significant, but someone needs to check the Don Rice losers, and early money in general at Tampa to verify. Haven't gotten around to that. I was checking on early money horses all over, and it looked good, till I noticed Don Rice had half the winners! So probably more of a trainer specific pattern, or maybe just Tampa, or maybe all the Rice horses get that action.

Two, it is very interesting that you mention 2 to 6 minutes to post, as in every angle I come up with, that is *the* predictive zone exactly as far as the toteboard is concerned. Which should be qualified by saying that I'm using whatever the racingchannel.com toteboard says as far as minutes to post. Often the actual race will be 1-2 minutes different. I sometimes make a bet at one minute, and turn on the track video to see that it is still 2 minutes to post (which is also sometimes delayed, so it's hard to tell).

Third, in trying to identify false favorites via the toteboard. It is obvious to everyone I assume that 3-1 favorites win less often than 3-5 favorites. So while the favorite averages about 33% wins, if you take the odds of the favorite into account, you pretty much know when you've got a "26%" percent favorite or a "40%" favorite. Other than that general indicator, I have yet to find any pattern that indicates a favorite is more/less likely to win by more than a percentage point or two. With one exception so far: favorites that get their action early, but not late win a dismal 16% of the time. I'm actually not sure how this charts on the odds line, because I came up with that by making a "favorite for the minute". For each minute, I looked at the NEW DOLLARS ONLY FOR THAT MINUTE for each horse. In other words, each time the toteboard is updated you'd subtract from the current win pool total for a horse his total from before the update, leaving you only the dollars wagered IN THAT MINUTE. In this way, we have a favorite for the 3rd minute before post, a favorite for the 4th minute, etc. Overall favorites that got their money before 8,9 or farther minutes out, who never appeared as the "minute favorite" at any point during the 2-6 minute range didn't win, even if they were still the overall favorites. If they are the minute fave just once in those crucial "near" minutes, they suddenly jump back up to at least a 29% percent win rate.

I see I'm going to have to write a glossary for all the new jargon this is going to require, or it's going to be more confusing than curling...

[stuball]

Hi GT
I totally agree with your concept of minute favs. It is impossible to track something like this without a computer..if you need someone to test your program (ideas,etc) contact me..!!!
I am totally blown away by what you have accomplished and would like to share in the research on this project...How's the parser doing?

Have a nice day !!!!!!!!

Stuball

[gumbo]

Interesting observations, GT!

For the record, Don Rice is the premier "local" Trainer at Tampa Bay Downs. On the plus side, his horses are extremely well-prepared (328 races in past year - $0.36 R.O.I)), and he is especially good with Maidens (MSW @ Tampa this meet 20 starts +$3.00 R.O.I).

On the minus side, his mounts often get overbet due to his reputation. Tampa is not a big simulcast track, so the locals have a bigger influence (I am not a "Homer", but anybody that is not giving the Tampa lines a good look is missing some excellent opportunities).

While I suspect my research is not as well-documented as GTs, I coiuld not agree more with the "2-6" minute observation. I think there is a logical explanation.

For whimsical sake, let's divide the betting public into 3 classes:

1. The numbers/favorite/it's a nice day crowd - These folks are happy to be at the Track - here Florida many are retired and are looking for a nice afternoon. They almost always bet before the 10-minute mark (to avoid the crowds), and typically bet the FAVE or somebody close. No handicapping (to speak of) involved.

2. The "absolute last minute" crowd - Some of us may fit into this, because we all watch the Toteboard as long as we can, but these are the folks who have little confidence in their handicapping and have "picked" about four horses and are simply waiting to see who goes off the longest.

3. Most of us - We have done our best to handicap the race. We must wait until the pool builds up to see if our selections have any value, but we usually know this with 3-4 minutes to go. We like to play the Toteboard, but we ain't gonna let it pick our horse. Occasionally, we get screwed, but we accept it as the nature of the beast.

We are the ones betting during those crucial 2 to 6 minutes before post.

gumbo

[Rick]

GameTheory,

One way of approaching this was discussed in the book "The Theory of Gambling and Statistical Logic" by Richard Epstein. He calculates what he calls the "entropy" of the race as follows:

First calculate the value p * log p for all horses in the race, where p = 1 / (odds + 1).

Then sum the values for all horses and take the negative of that.

He says that in races with relatively high entropy, there are often overlays in the 6-1 to 8-1 odds range. A high entropy race would be one in which there is no heavy favorite, the maximum value occurring when all horses have the same odds.

My own limited experiments with this many years ago found overlays at 8-1 (8.0 to 8.9). I also found that high entropy races were usually those with at least 9 horses and no horse at odds less than 3-2 so I just used those two simple rules for testing. I haven't checked lately but it might be worth a try. It did influence my thinking about what are the best kinds of races to find overlays in, and that has proven to be valuable.

[GameTheory]

That's just the sort of formula I was looking for... I'll check it out, thanks...

[GameTheory]

Rick, and any math wizards:

I know with a handle like "GameTheory" I ought to know this stuff, but I really don't know much about mathmatical regression, calculating variance, etc. Many of the formulas I run across use the log of this or that (like the one you just gave, Rick) -- am I to assume this is the "natural logarithm", or the common base10 log, or what?

[Rick]

GameTheory,

They usually mean natural logarithm, which is sometimes abbreviated ln in math books to distinguish it from the base 10 logarithm. But it usually doesn't matter which one you use because ln x = log x / log e, so ln x is just a constant (1 / log e = 2.302585) times the log x. In an equation like the one I gave, multiplication by a constant will change the values of the entropy measure but not the relative ranking of different races. So, use either one you want. I'd probably use natural logs, but that's due to my seeing it so often in math books and technical papers. Most people are probably more familiar with base 10 logs though. I guess it would depend on your audience if you're going to quote actual numbers.

[Rick]

GameTheory,

I was just looking at an old study that showed that 10 minute favorites that were not post time favorites showed a 32% loss. Conversely, post time favorites that were not 10 minute favorites showed only a 9% loss. Does your data show something similar to that? If so, it would appear that it would be better to look for overlays as early as 10 minutes rather than waiting until near post time.

[GameTheory]

Originally posted by Rick
I was just looking at an old study that showed that 10 minute favorites that were not post time favorites showed a 32% loss. Conversely, post time favorites that were not 10 minute favorites showed only a 9% loss. Does your data show something similar to that? If so, it would appear that it would be better to look for overlays as early as 10 minutes rather than waiting until near post time.


I'm not sure I understand.

If you bet on all favorites, I believe you can expect to lose (RANT => Why do so many people spell this as "loose", btw? "Loose" is the opposite of "tight"; "lose" is the opposite of "win".) about 15%, or about the same amount as the takeout.

The first group you to refer to are in fact not favorites, right? They were the favorite at the 10 minute mark, but were not the final favorite. And since they were not favorites, they probably win less than favorites, and a 32% loss is no surprise.

And the second group were final favorites that became the favorite (or re-gained the title) within the last nine minutes, and showed a better than average ROI for favorites, although still a loss.

If that's what you're saying, yes, that sounds correct. The predictive value of the public's money gets stronger the closer to post, in general. Separating the new dollars by minute as I explained earlier shows something like this:

"MINUTE" FAVORITES
(new dollars each minute only -- a single minute does not affect any other minute)

[min / win% / roi ]
min 1: 31.8% -14.9%
min 2: 31.4% -13.7%
min 3: 30.0% -16.0%
min 4: 27.2% -23.6%
min 5: 26.9% -21.7%
min 6: 26.8% -18.9%
min 7: 26.9% -16.4%
min 8: 26.5% -12.9%
min 9: 23.8% -22.8%
min 10: 26.3% -14.0%
(if you keep going it gets down to about 22% wins at min 17 or so.)
This was taken from a sample of appox. 1300 races at all tracks.


What I don't understand is the looking for overlays at 10 minutes part. What do you mean by that -- in relation to some odds line you have, or just as function of the toteboard pattern, or what?

[Rick]

GameTheory,

What I was thinking of was comparing my odds line to toteboard odds. If my minimum odds for an overlay were 3-1 and I made the bet at 10 minutes to post, should I regret making the bet if the odds drop later, or does it just improve my chances without affecting the value of the overlay?

Although you can't answer this question directly using the data, it does seem that money bet after 10 minutes is "smarter" than money bet before that. Of course, the best thing to do would be to use late money to adjust my odds line, but that's hard to do manually.

The loss rates at less than 10 minutes on favorites overall seem to be less than 20% which would seem to support the theory (I assume that favorites lose about 20% for all time periods combined). What I'm trying to do is get an overlay based on "dumb" money and assume that "smart" money won't hurt me.

What I'm curious about is whether becoming a "new" favorite is significant as you get closer to post time. For example, the horse wasn't the favorite at 3 minutes to post but now is at 2 minutes to post. If the ROI is better when a horse is bet down like this maybe I should consider the drop in odds to be a positive thing. It's possible that we may be hurting ourselves by waiting until too late and eliminating some of our best bets. I firmly believe that the only truly accurate odds line would have to use toteboard odds as one of the inputs.

[GameTheory]

Well,

One thing I would like my toteboard program to do is to predict the final odds at the 2 minute mark. Most of us have an intuitive feel for this after a while, but teaching the computer is another story.

Your entropy formula is of some help there. If you look at the changes in the "entropy score" from 10 minutes, that says a lot. If it goes up during that time, the public is unsure of themselves and you usually see the winner come from the 3rd or 4th or 5th public choice. If that number goes down, the winner usually comes from the 1st or 2nd choice. Again, this is very easy for a person to tell watching the board, but having a single number is nice to feed the computer.

As to whether the public's money is really smarter closer to post I'm not sure about. Statisically, it is, but is that because smarter people are betting late? I doubt it. I think it has more to do with the fact that more people are betting closer to post, so it represents a truer consensus of public opinion. The bigger the track, the more stable the odds lines are. I was trying to think of way to estimate the number of PEOPLE (in a relative way) betting on each choice as opposed to the number of dollars, but it probably isn't possible. You would have to do a lot of comparison between the pools I think, and somehow try to identify large bets and lessen the weight they have.

Anyway, is it just as good comparing your odds line to the toteboard at 10 minutes than at 2 minutes? I don't think so. Actually, I better than taking any single snapshot at a single time is to track the changes over the period, in which case you will usually have a very good idea of the final odds and can act accordingly. Personally, I think the whole comparing your personal odds line to the toteboard method is flawed, unless used in only a very general way. A much more effective method (I think) is to handicap the race as the public does, looking at the more obvious factors for that type of race, and then handicapping it as you normally do. You should have a sense of which horses are going to be overlays hours before the first post. And if the odds are behaving differently than you expect, stay away from that race -- you haven't got it. I think one of the worst bets around is a horse that LOOKS (to the public) like it should be 2-1 and is going off at 5-1. That's not an overlay, it's a trap.

[Rick]

GameTheory,

I agree with you. We should be able to predict "dumb money" in advance based on obvious past performance factors. However, so far I haven't been motivated enough to research that so I currently rely on things like early odds and morning line. If you know of anyone who has done any research in this area I would be thankful if you would point me in the right direction. Nevertheless, thanks for your generous contributions.

[GameTheory]

Predicting the public is basically an intuitive thing -- nothing too scientific. Just look for the horses that jump out at you with the obvious factors -- big figs, lots of 1's in the PP position lines, good connections, etc. The trick to predicting the public is NOT to look very deeply into things, so it is usually a good idea to do it before you do your own handicapping or you get biased by your deeper analysis.

Pizzolla in HANDICAPPING MAGIC says pretty much the same thing with his "soft focus" technique -- that's a pretty good book by the way...