8-1 and up longshots

[dav4463]

When I review my records, my major profits occur when 8-1 and up longshots either win or hit the board keying an exotic payoff. Almost everybody I talk to insist that you cannot consistently profit playing these longshots. Records don't lie, but in the back of my mind I wonder if I am sitting on one of those 0 for 100 streaks that longshot players are warned against. My question is are there any other players out there that consistently profit on longshots only ? When I play less than 8-1 horses, I feel like I'm treading water, I win some, lose some, and never really get ahead until one of these longshots come in. So, would I be better off just playing the longshots only ?

[Jeff P]

Almost everybody I talk to insist that you cannot consistently profit playing these longshots.

Who in the world are you talking to? I doubt that these people are winning players.

My own records parallel yours. They indicate that it's only horses winning at a decent price that make coming out ahead in the long run possible.

Decent price can be a moving target, depending on the factors you are using in making your selections. Some horses stand a much better chance of winning than others. In some situations, a horse might be 60-70 percent likely to win. In that situation, even money can be a decent price. In another situation, a horse might be 15-18 percent likely to win. In that situation, 8-1 could be considered a decent price.

I purposely gave two different examples at two different ends of the spectrum. Overall, my own records tell me that I've had far more success and should prefer betting what I perceive to be the lower percentage chance to win horse at the longer price over the logical horse at the shorter price. It would appear that the longer priced example allows me to have more room for error on my part when estimating a horse's chances of winning. In those situations when I am right the longer odds tend to more than compensate me for those that I was wrong about.

Another reason I like the longer priced horse over the shorter priced variety is late money. Shorter priced horses tend to drop in odds after the bell. Longer priced horses tend to go up in odds after the bell.

I've seen many many horses that were 3-1 at the gate win and end up paying $5.80. Everybody has. And everybody complains about it.

Call me crazy but I've also noticed lots of horses that were 10-1 or 12-1 at the gate end up paying $34.00 when they won. Everybody complains about the lower priced horse dropping in odds, but honestly, with longer priced horses I am very often pleasantly surprised by just the opposite.

[Fastracehorse]

I've posted 129 longshots this year - min'm price 6-1 - :(

I think that's evidence that they can be caught.

fffastt

[hurrikane]

I wouldn't play under 6-1. These 'longshots' are the way to make money. I have to be able to see what the public doesn't and capitalize on it. It's the only I can win.

Anything else and I would end up like formula2000

[sjk]

Earlier this year I expanded my play to include any horse that I give at least a 4% chance to win (formerly used 7.5%). That is not the only change I have made this year so it is not possible to isolate the effects of this change, but my overall results have been improved and I believe playing these longer odds horses has been a positive.

In my case I don't restrict my play to the longer prices; I will take value wherever I can find it including $12.00 exactas if I think I'm getting value.

I made the change to my betting parameters only after significant back-testing gave me good reason to believe that it would be a benefit. I see that during the last 4 months my win pct is below 11% (used to be 12%+), but the higher prices and the expanded betting opportunities more than compensate.

You need to be able to handle the losing streaks. I lost 32 in a row on Sunday and 0-40s are pretty common.

[Valuist]

No question the price shots will help out most in the long run. The horses that were 8-1 and up probably had some fundamental reason that you liked them. Now lets say you change your strategy and force the issue: you only bet 8-1 shots and up. Now instead of only playing 8-1 and up value horses you could end up forcing some horses that deserve to be big odds. A good price is only good if the horse can make it pay off. And psychologically, going too long between trips to the pay window can be very hazardous. Also, it is possible for a horse to be 3-1 or 4-1 and be worth a wager as well.

I used to track every kind of wager I made and found patterns that were supposedly more profitable than others. I had a very profitable ROI on route races but was losing money in sprints. Easy solution, just drop sprints and bet routes only, right? Wrong. Enough horses switch back and forth that to truly be sharp, I needed to follow both sprints and routes in the circuits I play. A synergistic effect, I guess. Good luck if you do make a big change in your strategy. Not saying it can't work but don't be discouraged if the results don't turn around quickly right away.

[JustMissed]

Of all the U.S. t-bred races ran yesterday, 9/1/2004, does anyone know how many horses won at 8-1 or better?

JM

[BillW]

Originally posted by JustMissed
Of all the U.S. t-bred races ran yesterday, 9/1/2004, does anyone know how many horses won at 8-1 or better?

JM

23 out of 115. I don't understand why that is important?

[JustMissed]

Originally posted by BillW
23 out of 115. I don't understand why that is important?

Thanks for taking the time to look that up.

I was curious as to how many plays a player would get, on average, if he limited himself to 8-1 and higher horses.

I am surprised at 20%. I would have expected a lesser number.

Thanks,

JM

[Skanoochies]

Wow thats incredible. Bet $2 to win on all 115 and even if the 23 winners were at the minimum 8 to 1 you show a profit of $184. Bet that wouldn`t happen too often.

Skanoochies.:)

[Skanoochies]

Oops did you mean 23 winners out of 115 races not runners? Mea culpa.:o

[BillW]

Originally posted by JustMissed
Thanks for taking the time to look that up.

I was curious as to how many plays a player would get, on average, if he limited himself to 8-1 and higher horses.

I am surprised at 20%. I would have expected a lesser number.

Thanks,

JM

I'm sure that is not average, but the plays aren't that sparse and remember a lot of plays don't win :). Selectivity is (as with all ranges) a must.


Bill

[JustMissed]

Originally posted by BillW
I'm sure that is not average, but the plays aren't that sparse and remember a lot of plays don't win :). Selectivity is (as with all ranges) a must.


Bill

Ditto that Bill.

I am always interested in other players betting plans and staking methods.

We had a very interesting discussion with a long odds player here a few weeks back, can't remember his name. He was playing with a large bankroll, very high odds and fairly long loosing streaks with low strike rate.

If I remember though his payout when he hit was so high and he bet so much his P.O.T. was serious money.


JM

[Fastracehorse]

1) I avoid long losing streaks by spot playing.

2) I don't bet for sake of value - I think that is counter-productive.

I bet for sake of high probabilty/value. Maybe that is what people are saying.

fffastt

[Zaf]

Originally posted by BillW
I'm sure that is not average, but the plays aren't that sparse and remember a lot of plays don't win :). Selectivity is (as with all ranges) a must.


Bill

I agree Bill, That reminds me how selective I was last week at Saratoga :eek: :eek: :eek: :eek:

Anyway I agree with that wholeheartly, I felt like a kid at Disney Land last week :)

ZAFONIC

[Dutch]

I am more comfortable betting the longer priced horses, for some of the reasons above. Of course, you want to bet when you think it has a better chance than the odds show. If I thought it was really a 10-1 horse, the probability is pretty low that it wins.

Some of the theory that I bought into before, was to bet more when you think your edge is greater. Edge being defined as the difference between the odds and the probability you give it.
Example: You think it should be 3-1, it goes off at 4-1, doesn't merit much of a bet. You think it should be 2-1, it goes off at 4-1, the edge is better, so you bet more. Theory being you will eventually win more when the edge is higher

Most research shows the probability of really high horses - 39-1 or more - winning, is so low, they don't merit a bet.

I've heard a lot of players say they restrict betting to 5-1 or some similar figure, but who knows how it works out?

[Fastracehorse]

Most research shows the probability of really high horses - 39-1 or more - winning, is so low, they don't merit a bet.

I've heard a lot of players say they restrict betting to 5-1 or some similar figure, but who knows how it works out?

==================================

I don't agree with that - I've had some bang-up weeks with 39-1's or greater.

Last Wednesday I bet a 50-1 firster at Wdb - 2nd a nose.

Tough beat really, just nailed at the wire. Paid $37 to place though.

I bet the horse with the other fave on top, oh well.

fffastt

[kenwoodallpromos]

Sunday I bet 3 at 10-1+ in a race with 5 entries. Won $33.40.

[linrom1]

Many of us wish that there were many more long shot players. The truth is that there are so few of them, that it makes this game so though for the rest of us.

[BillW]

Originally posted by Skanoochies
Oops did you mean 23 winners out of 115 races not runners? Mea culpa.:o

Yep, 23 races won by long shots (defined as > 8-1) out of 115.

BTW this was followed by 23 out of 164 yesterday. Coming back to reality a bit from the high pctg. the day before.

They are there (My brother caught the $57 winner at CT last night) but you have to sniff them out :).

Bill

My longshot philosophy: I hate it when others agree with me as I find they are usually wrong :D

[Equineer]

Races Horses Wins
All 77137 647742
Under 8.00/1 306147 47.26% 63095 81.80%
8.00/1 & Up 341595 52.74% 14042 18.20%Excluded races: dead heats, coupled entries, betting fields.

[linrom1]

I bet your data is skewed in favor of long shots. Have you looked at it for major tracks only?

[Niko]

[QUOTE]Originally posted by linrom1
[B]Many of us wish that there were many more long shot players. The truth is that there are so few of them, that it makes this game so though for the rest of us.

I'm interested in your thought process behind this....


My ROI is higher on longshots. I USUALLY can't discern enough of a difference on lower odds horses to be profitable.

[linrom1]

I'm interested in your thought process behind this...


I want better odds for horse that can actually win.

[Fastracehorse]

I think your handciapping is skewed towards favorties.

Longshots play a significant role at the track.

fffastt

[Niko]

Linman,
you're one of the very few that I'm aware of that can make a profit just playing low odds horses.

Rebates, full time occupation or a combination?

[Niko]

linrom.....SORRY

[msola1]

A couple of posts on this thread say things like "if a horse has a 60% chance to win," or "I only use horses that have a 4% chance to win."

My question for anybody who can answer is how do you determine these percentage "chances to win"? How do you know a horse has a 4% chance? If you are going to say that you do it by feel, forget it. I'm looking for ways to express numerical evaluations as percentages or odds.

Here's an example of what I'm grappling with: Let's say I have an algorithm that gives for each horse in the race a number, lowest being best. And let's say that for two races I have the following ratings:

Race 1: 1__4__ 5__15
Race 2: 1__2__10__15

I think it is intuitive that in race 1, the low-rated horse's odds would be lower than the low-rated horse in race 2, because he's relatively better than the whole field.

Any idea on how to convert these numbers into percentages or odds?

[BillW]

Originally posted by msola1
A couple of posts on this thread say things like "if a horse has a 60% chance to win," or "I only use horses that have a 4% chance to win."

My question for anybody who can answer is how do you determine these percentage "chances to win"? How do you know a horse has a 4% chance? If you are going to say that you do it by feel, forget it. I'm looking for ways to express numerical evaluations as percentages or odds.

Here's an example of what I'm grappling with: Let's say I have an algorithm that gives for each horse in the race a number, lowest being best. And let's say that for two races I have the following ratings:

Race 1: 1__4__ 5__15
Race 2: 1__2__10__15

I think it is intuitive that in race 1, the low-rated horse's odds would be lower than the low-rated horse in race 2, because he's relatively better than the whole field.

Any idea on how to convert these numbers into percentages or odds?

msola1,

Welcome to the site.

The most direct way would be to divide the rating by the total of all ratings and multiply by 100.

Race 1: horse rated 1 would be 1 divided by 1+4+5+15 or 1 divided by 25 = 0.04 times 100 = 4%

horse rated 4 would be 4/25 x 100 = 16% etc.

Bill

[sjk]

msola1,

In my case I use spent months, or maybe it was years working on a program to calculate these percentages.

The simple answer to your question is: 1). assemble a large database of past races; 2). identify the parameters that you want to use to predict you win percentages; 3). calcuate your parameters for those races; 4). carefully evaluate how the actual race results relate to your parameters. You probably need to look at a number of parameters and combining the effects of numerous parameters will take some careful work.

None of this is quick or easy but it my case it has led to good results.

[Equineer]

Originally posted by linrom1
I bet your data is skewed in favor of long shots. Have you looked at it for major tracks only? Compared to races run for ALL meets, it may be skewed towards the major tracks: APX, AQU, CDX, BEL, GPX, HOL, SAX.

Among 62 tracks in study, were 3-years-full-meets for AQU, BEL, GPX, HOL, and SAX, plus good representation for APX, CDX, KEE, DMR, SAR. Even if you don't play every major card, you pretty much have to download/handicap the majors to get decent samples for top echelon races.

But... and a big but, if you mean skewed towards minors because Typical-Joe plays mostly majors... then you are right because betting activity rather than races becomes your distribution curve, and cheap large fields, common at many minor tracks, do include more longshots to handicap.

[formula_2002]

from another subject, but it seems to fit here.

"The probability of showing a profit (without handicapping) in a series of 100 plays increase as the dollar odds increase.

odds
1-1, 0 wins in 21 sets (0% probability of profit)

10-1, 15 wins in 86 sets (14% probability)

10 to 20-1 odds , 15 wins in 86 sets (17% probability)

>20-1 odds, 24 wins in 104 sets (24% probability)

conclusion:
It is easier for some one who limits his plays to higher odds, to think he can beat the game !!"

Joe M

[msola1]

Bill,

Thanks for your answer. The problem is a little different, howver. In my case the low number is the most likely winner, not the least likely. Simply inverting and then doing what you did won;t work, and isn't valid math probabilistically--try it.

Mike

[BillW]

Originally posted by msola1
Bill,

Thanks for your answer. The problem is a little different, howver. In my case the low number is the most likely winner, not the least likely. Simply inverting and then doing what you did won;t work, and isn't valid math probabilistically--try it.

Mike

Mike,

I suspected that might be the case. For this case subtract each rating from 1 plus the largest individual rating (15 in both cases you cite) and follow the above procedure.

the horse rated 1 would be (15+1) - 1 or 15
the horse rated 4 would be 16 - 4 or 12
the horse rated 5 would be 16-5 or 11
the horse rated 15 would be 16-15 or 1

now your total is 39 and the horse originally rated 1 is 15/39x100 etc.

Is that what you are looking for?

HTH,

Bill

[Equineer]

SJK,In my case I use spent months, or maybe it was years working on a program to calculate these percentages.In another thread you explained that you are now systematically trusting your handicapped probabilities down to the 4% level... with profitable results. That tells me you are too modest about either/both your skill and/or work ethic.

Re-emphasizing that you are systematically doing this... you really are a true Value Pro!

[sjk]

Equineer,

Appreciate the kind words. SJK

[Jeff P]

Originally posted by msola1...

A couple of posts on this thread say things like "if a horse has a 60% chance to win," or "I only use horses that have a 4% chance to win."

My question for anybody who can answer is how do you determine these percentage "chances to win"? How do you know a horse has a 4% chance? If you are going to say that you do it by feel, forget it. I'm looking for ways to express numerical evaluations as percentages or odds.



Warn ya ahead of time. This post is not going to be short.

My own probability estimates for a horse's chances of winning are arrived at using my own methodology that I'll try and describe below.

First, I've developed (what I think anyway) to be some very good software and a fair sized racing database to go with it. This is a pet project of mine that I've been dedicated to for about the past 20 years now. Along the way I've created some very good query tools for getting information out of my database and displaying it in front of me in an easy to understand format that tells me very quickly what I want to know.

I'm a bit of a free thinker and have a tendency to look at races in my own way rather than someone else's. So a lot of unique numbers and compound ratings have made it into my software. Most are entirely of my own creation. Some I've bastardized from concepts I've read about. And some were given to me by other experienced handicappers who've been kind enough to share/swap valid information with me from time to time.

One of those unique numbers I've come up with is a comprehensive rating. I call it my JRating simply because my first name is Jeff. This JRating employs a myriad of factors. Each separate factor used in the JRating is weighted according to the true impact, both in terms of win probability and money returned, that the individual factor showed during some fairly significant testing.

Each separate factor used in the JRating calculation falls under what I've heard referred to as an 'umbrella' or 'school of handicapping.' I've got some very good compound factors that represent pace or early speed. That's one school. I've got some very good unique numbers that represent class. That's another school. I've got some very good compound numbers that represent overall ability. Is that the same thing as class? Maybe. But let's save that for another discussion. I've got a very good compound number for form. There's another school- and very few others have attempted to quantify that. I started a thread on my form rating in the software forum back in May and posted some benchmark testing results at the time. I've got some very good ratings for jockeys and trainers. There's another school. So what I did was code out an algorithm to assemble all of these things together into a single number, my JRating.

My initial goal was to create my own unique number similar in predictive strength to the Bris prime power rating. I figured that if it was unique and nobody else could see it then it might return more money from a $2.00 win bet than the Bris prime power number while still being very predictive in nature. What I hate about prime power is that you'll lose significant money if you bet $2.00 to win on every top rated prime power horse. What I like about prime power is that it does have some predictive strength.

Initial testing of my own JRating revealed that it could be every bit as predictive as prime power, but like prime power it just wasn't profitable on its own. So I began looking at different segments of horses, trying to see if by excluding them I could raise the roi. What I found was that by doing some tweaking, and elevating scores in certain areas for horses that looked bad on paper, and lowering scores in certain areas for horses that looked good on paper, that I could improve the overall roi of the JRating without cutting back on its predictive strength too much.

After some experimenting, I arrived at a place where the top J-rated horse consistently wins about 25 percent of all dirt races and a $2.00 win bet on every top J-rated horse shows a small profit. These numbers that I am quoting here are from testing on fresh data, not data that I used while developing the algorithm.

Okay. That's a synopsis of my general process. Now to answer your question. How do I estimate a horse's chances of winning a given race?

One of the things I get from my software are results whenever I run a query. One of the screens I created shows a breakout of how all horses ranked on a selected factor have performed historically. The screen displays the following information on each line: Min Rank, Max Rank, Wins, Plays, Percentage, and Roi. Each screen has 20 lines. So from a single query based on a single factor, I get to see how the top ranked horse for that factor performed, how the second ranked horse for that factor performed, how the third ranked horse performed... all the way down to the twentieth ranked horse for that factor.

So when I run a query based on JRating, I can see that the top ranked horse has approximately a 25-26 perent chance of winning, the second ranked horse has a 17-19 percent chance of winning, the third ranked horse has a 13-14 percent chance of winning, the fourth ranked horse has a 12-13 percent chance of winning... all the way down to the twentieth position. These percentages are really just historical probabilities based on thousands of occurances.

Now here's where things start to get interesting. This is really just a starting point, isn't it? You see not all top J-rated horses are the same. I also have the ability to run complex queries combining JRating- not just for rank but using numeric value and gap definitions as well- with numeric value, rank, and gap definitions for as many other factors in combination with each other as I see fit. And indeed some of the queries I have put together do reveal situations where it could be said from historical probabilities that a horse had a 60-70 percent chance of winning its race.

Is this an accurate way of arriving at probability estimates? Auto insurance companies determine our rates when actuaries classify drivers using historical probabilities based on millions of occurances. They make lots of money. Sure, they have a lot more data than I do. All I've tried to do is adopt their overall process as much as possible and apply it to thoroughbred horse racing.

[formula_2002]

Jeff

How do you weigh each stand alone factor and how do you determine which factors to combine ?


Joe M

[Equineer]

Jeff,

True probability handicapping isn't mainstream handicapping, and some well-respected authors/developers have dismissed it as a futile pursuit, sort of like alchemy.

I'm glad to see you (and others like SJK) are proving the naysayers wrong.

That said, I would like to add another consideration to the requirements for good probability handicapping.

Average accuracy and routine usefulness do not necessarily go hand in hand.

Let's say exhaustive testing demonstrates apparent accuracy. For example, 20,000 horses that you project at 10% probability actually win at a 10% rate and return a $2-Net of $1.87 (versus $1.6x, where x is dictated by actual takeouts/breakage).

You may be pleased with the average accuracy, and you may conclude that value opportunities must be exist. But you can also be misled if you jump to the conclusion that your results are systematically useful (in a practical sense, race after race).

For those same 20,000 horses that won 2,000 races, you need to study the distribution of actual track-odds-equivalent probabilities. From this, you hope to find that actual parimutuel probabilities (public odds if you will) reflect a symmetric but reasonably narrow distribution peaking at or very near 10% for the same horses that you pegged at 10%.

Especially for the subset of 2,000 winners, you don't want to find that central tendency is skewed towards pari-mutuel probabilities below 10%, such that your $2-Net of $1.87 was inordinately dependent on a few extreme outliers among high-priced winners.

This sort of distribution analysis, applied comprehensively, tells you whether your probability projections are likely to be systematically useful for evaluating overlays and underlays.

[sjk]

You could make the prediction that every horse in a 10 horse field has a 10% chance to win and be correct on the average. That is not going to help you make money.

I always look at the horses that I would have bet because they were overlaid by a chosen amount (I choose the amount to maximize my results).

Unless your line is a lot better than mine you will find that the universe of horses that you expect to win 10% of the time which are also overlays do not actually win 10% of the time.

Hopefully you have been sufficiently accurate in making your line to give you net positive results because the extra price makes up for the lost win pct.

[msola1]

Bill,

That's still not it. I think it needs much more sophisticated math.

If you look at both examples, and do what you suggest, the top horse will in both cases be 1.6/1. This is counterintuitive. If the top horse is relatively further away from the rest of the field in ratings, his odds should be relatively lower than when he is closer to other horses in the field.

I went through this process already. Here's another consideration that is confounding me: You would think it is necessarily true that if you add the percentage probability of all the horses in every race, the sum should always be 100. But if you look at any array of odds in a race and convert those odds back into probability, you will find that it adds up to usually more than 100%.

Mike

[Jeff P]

posted by Joe M...

How do you weigh each stand alone factor and how do you determine which factors to combine ?


Joe,

Through lots and lots of testing, one factor at a time.

My software uses the $1.00 bris drf single format data files. So a very simple thing to look at is the effect of the last race bris speed figure. Running a query based on rank for last race bris speed figure I can see that in dirt sprints the top rated horse on that factor won approximately 27 percent of its races. If you had bet every horse in the sample with the top last race bris speed figure to win you'd have lost 14 cents on the dollar with an roi of .8608.

Now, let's compare that to all horses in the sample. If you'd have bet every horse in the sample to win you'd have cashed 12.25 percent of your tickets and lost 22 cents on the dollar with an roi of .7804.

So the horse with the best last race bris speed figure has a win percent based impact value of approximately 2.20 and an roi based impact value of approximately 1.10.

Conclusion: betting the horse with the best last race bris speed figure to win is better than betting all horses to win. Or put another way, and this is more in line with the way that I think: not betting horses ranked poorly in last race bris speed figure is far better than betting all horses.

Not all factors are created equal. For whatever reasons, the betting public historically likes some factors while ignoring others.
Thankfully, there are factors with roi based impact values that outperform the last race bris speed figure.

[Equineer]

Here is an example from another thread that probably is more germane to this thread.

ON THE RIGHT SIDE: how software handicapped horses from past performances for 77K races, showing how the odds and probabilities calculated by the software compared to actual win percentages.

ON THE LEFT SIDE: how the public handicapped the same 77K races, showing how odds and probabilities produced by wagering compared to actual win percentages.

ODDS/PROBABILITY COMPARISONS SECTION 2 of 161

ALL-RACES_(All Categories)_[62 Tracks, 77137 Races, 647742 Horses]

TRACK_ODDS_LEVEL_STATS____________ HANDICAPPED_ODDS_LEVEL_STATS______
ODDS Odds Win $2 Odds Win $2
LEVL ____NH ___NW** __EW Prob _Pct _Net ____NH ___NW** __EW Prob _Pct _Net
0 43 33 31 72.1 76.7 1.68 145 84** 105 72.3 57.9 1.85
1/5 634 459* 399 62.9 72.4 1.83 569 325** 359 63.2 57.1 1.95
2/5 2240 1380* 1234 55.1 61.6 1.79 1746 947 961 55.0 54.2 1.96
3/5 4483 2381* 2184 48.7 53.1 1.75 3476 1690 1694 48.7 48.6 1.93
4/5 6313 2995* 2762 43.8 47.4 1.75 5223 2299 2280 43.7 44.0 1.87
Even 7422 3030 2939 39.6 40.8 1.67 6708 2612 2655 39.6 38.9 1.85
6/5 8297 3100 3000 36.2 37.4 1.67 7947 2788 2874 36.2 35.1 1.81
7/5 4461 1575 1514 33.9 35.3 1.69 4371 1502 1483 33.9 34.4 1.83
3/2 4536 1527 1479 32.6 33.7 1.68 4430 1404 1445 32.6 31.7 1.78
8/5 9613 3108* 2961 30.8 32.3 1.71 9401 2869 2896 30.8 30.5 1.80
9/5 10408 3096 2983 28.7 29.7 1.69 9702 2713 2783 28.7 28.0 1.75
2/1 28166 7418* 7214 25.6 26.3 1.68 24533 6175 6285 25.6 25.2 1.78
5/2 28185 6341 6260 22.2 22.5 1.66 25014 5574 5554 22.2 22.3 1.77
3/1 27170 5281 5323 19.6 19.4 1.62 24808 5037* 4859 19.6 20.3 1.78
7/2 25251 4370 4428 17.5 17.3 1.62 24032 4405* 4213 17.5 18.3 1.76
4/1 23173 3492** 3677 15.9 15.1 1.56 23451 3762 3718 15.9 16.0 1.69
9/2 20634 2962 2988 14.5 14.4 1.63 22338 3361 3234 14.5 15.0 1.75
5/1 35984 4445** 4622 12.8 12.4 1.58 42001 5707* 5391 12.8 13.6 1.74
6/1 31619 3436 3516 11.1 10.9 1.61 38013 4473* 4223 11.1 11.8 1.74
7/1 27515 2666 2698 9.8 9.7 1.63 34178 3340 3351 9.8 9.8 1.64
8/1 24382 2209 2138 8.8 9.1 1.70 30592 2796 2682 8.8 9.1 1.67
9/1 21788 1719 1728 7.9 7.9 1.64 27081 2160 2147 7.9 8.0 1.61
10/1 19574 1391 1418 7.2 7.1 1.62 24177 1758 1751 7.2 7.3 1.59
11/1 17384 1139 1158 6.7 6.6 1.62 21646 1417 1442 6.7 6.5 1.57
12/1 15376 939 948 6.2 6.1 1.63 19156 1130 1181 6.2 5.9 1.53
13/1 14102 771 809 5.7 5.5 1.57 17043 936 978 5.7 5.5 1.53
14/1 12506 644 671 5.4 5.1 1.59 15745 764** 845 5.4 4.9 1.46
15/1 48048 2056** 2191 4.6 4.3 1.55 58124 2486** 2656 4.6 4.3 1.44
20/1 33613 1102** 1199 3.6 3.3 1.52 35952 1107** 1285 3.6 3.1 1.25
25/1 85324 1721** 2027 2.4 2.0 1.39 62809 1285** 1544 2.5 2.0 0.98
50/1 49498 351** 590 1.2 0.7 0.94 23331 231** 276 1.2 1.0 0.60
Avg%Diff/Horse Avg%Diff/Horse
ALL: 647742 77137 0.472% 647742 77137 0.402%


ODDS LEVL: really an odds/probability range for both sides, where (for
example) 9/2 includes parimutuel odds from 4.50 through 4.95
NH: number of horses in each odds level for each side.
NW: actual number of winners in each odds level for each side.
EW: expected winners in each level, as predicted by the probabilities that
correspond to parimutuel odds in odds levels for each side.
ODDS PROB: probabilites that correpond to the odds levels for each side.
WIN PCT: actual win percentage for horses in each odds level in each side.

[Jeff P]

posted by Equineer...

Average accuracy and routine usefulness do not necessarily go hand in hand.

Absolutely agree with that statement. I've found that the best factors I use to derive my probability estimates have to deviate in some way from what the public is thinking in order to produce consistent overlays.

The public likes classy horses that finished well in their most recent starts that have good speed figures relative to the field that are handled by high percentage trainers and name jockeys. I realize I'm going to an extreme here by describing a recipe for an overbet favorite.

My point is If I used that as my selection criteria, yes, I could establish a pretty good win rate. But quite obviously it wouldn't be useful.


For those same 20,000 horses that won 2,000 races, you need to study the distribution of actual track-odds-equivalent probabilities. From this, you hope to find that actual parimutuel probabilities (public odds if you will) reflect a symmetric but reasonably narrow distribution peaking at or very near 10% for the same horses that you pegged at 10%.

One of the things I do look at are factors and combinations of factors in terms of odds range distributions. And this gets back to the heart of the question posed in the very first post of this thread. I find that it's primarily not at the lower end of the odds spectrum where true overlay opportunities exist.

I find that horses ranked tops in the field for certain factors, such as some of my compound pace ratings or my JRating, when they are over 9-1 in odds, are quite profitable when taken as a group. If you look at win percentages at each odds range you'd notice that as the odds go up the win percentage goes down. At above 20-1 the win percentage might be very low.

But there always seem to be a few $70.00 to $100.00 horses in that range above 20-1 whenever I look at a large enough sample size. You'd hate to have to rely on finding bombs like that. But every large chunk of data that I look at has them. So I never let really long odds scare me off.

I'm not really shooting for hitting a 10 percent win rate at horses I pegged at 10 percent. I do know that I'm in the ballpark though. At the higher end of the odds spectrum the occasional windfall paid by a bomb is enough to allow some room for error.

[formula_2002]

Jeff, "one factor at a time".

I have also tested one factor at a time. I Use the all-ways data files and have tested each factor (about 80 +) by incremental odds line .

Some do better then others, some do better then the public and some even show a profit in the >=6 and <20-1 odds ranges.

(the same is true of the 100's of combinations I have tested)

But I have yet to make any of the "profits" repeat with any confidence.

My analysis of this data has cause me to cut my betting to "pass time" level only.
I'm saving a fortune!!
Joe M

[sjk]

Joe,

Perhaps should should try to create your own factors rather than relying on those that are publicly available.

Also, the reason for testing each "factor" individually is to maximize the definition and the application of that factor. I would never expect any one factor in isolation to be of any use whatsoever. The competition (other players) is using all of the information so it it necessary to use all of the information to have a chance against them.

I hesitate to use the word factor since that carries the implication that "factors" will be put together in some linear sort of way. As I said above, the way you fit the factors together will take as much care as the development of the factors themselves.

[BillW]

Originally posted by msola1
Bill,

That's still not it. I think it needs much more sophisticated math.

If you look at both examples, and do what you suggest, the top horse will in both cases be 1.6/1. This is counterintuitive. If the top horse is relatively further away from the rest of the field in ratings, his odds should be relatively lower than when he is closer to other horses in the field.

I went through this process already. Here's another consideration that is confounding me: You would think it is necessarily true that if you add the percentage probability of all the horses in every race, the sum should always be 100. But if you look at any array of odds in a race and convert those odds back into probability, you will find that it adds up to usually more than 100%.

Mike

Mike,

Looks like you are looking for something different than the simple realtionship between numbers. Usually when rating contenders, less than 100% is allocated and the probabilities are scaled to 80% or what ever your allocation is. This may provide a closer result to what you are looking for.

As far as the odds adding up to more than 100% in races you have to take into account the takeout and breakage which inflates the probability of a horse (i.e. drops the odds paid) because odds are being set based on approximately 80% (17%-18% takeout, usually plus any breakage ) of the pool rather than 100% which would yield the correct corellation you seek. Usually they add up to greater than 120%.

Bill