Eliminating Non-Contenders? Ideas? Math?

[podonne]

I'm very curious to see how various people\methods go about "eliminating the non-contenders". Every handicapping book i read says this is the first thing to do, but I can't find any good statistics to back up the "easy eliminations" they recommend.

So I built this little number to try out a few ideas I have, but I don't have that many. Anyone have some ideas they've always wanted to test? I'll be glad to share the results here of every idea, or send to you directly if you prefer.

I have a database of 898 Fast Dirt races from Aquaduct's 2004 season.

I'll post a few of my tests here over the next few days...

-Phil O'Donnell

Test Name:
"Eliminate the horse with the lowest BRIS Power Rating if the last race was at a lower class level than this race"
Result:
432 Horses eliminated in 432 Races
6 Eliminations per Hundred finished 1st (94% Accurracy)

[Secretariat]

I'm very curious to see how various people\methods go about "eliminating the non-contenders". Every handicapping book i read says this is the first thing to do, but I can't find any good statistics to back up the "easy eliminations" they recommend.

So I built this little number to try out a few ideas I have, but I don't have that many. Anyone have some ideas they've always wanted to test? I'll be glad to share the results here of every idea, or send to you directly if you prefer.

I have a database of 898 Fast Dirt races from Aquaduct's 2004 season.

I'll post a few of my tests here over the next few days...

-Phil O'Donnell

Test Name:
"Eliminate the horse with the lowest BRIS Power Rating if the last race was at a lower class level than this race"
Result:
432 Horses eliminated in 432 Races
6 Eliminations per Hundred finished 1st (94% Accurracy)

Good post, but I suggest factoring in ROI into your tests. I agree about some of these non-elim rules published. Some of my tests have shown these elim tests could actually be wokred into longshot methods. Low hit rates but boxcar mutuels.

[valueguy]

This opinion will go againest everything you read on handicapping,but a quick and dirty way to elimnate horses is to look at the morning line if its over 8/1 on a horse its chances of winning are usually 50% less than what the M/L indicates .After you play a track for a while you will get a feel for accuracy of the odds maker.High M/L odds horses do win of course but its usually a shipper or a long lay off horse .The horses i like to watch for are M/L horses that are 5/1 or less but are going off at much higher odds .I take a close look at these .Sometimes you will find a good overlay.
FWIW

[podonne]

Test:
Eliminate all horses with morning line odds greated than 8/1
Result:
Eliminated 2456 Horses in 797 Races
11 Eliminations per 100 finished at least 1st (89% accuracy)
31 Eliminations per 100 finished at least 2nd (69% accuracy)
57 Eliminations per 100 finished at least 3rd (43% accuracy)
93 Eliminations per 100 finished at least 4th (7% accuracy)

[so.cal.fan]

How about eliminate all horses in the race that are not one of the first four favorites in the wagering when the horses leave the paddock.
My late father used this method successfully for many years in So. Calif.
He would handicap the horses, look at the horses in the paddock, live, circle horses who looked good, then check the tote board when they left the paddock. He would usually box contenders in exactas unless he was using the first three favorites....then he would pass the race.
He seemed to always have winning meets.

[spilparc]

1. When the morning line favorite is 3-1 or higher eliminate this horse.

2. When the morning line favorite is 3-1 or higher eliminate this horse and the next closest horse in odds. The second favorite.

[Tom]

All races since April 2005

# races - 61,536
#>8-1 ML - 193,116
# wins - 8213 4% of horses, 13 percent of races

[podonne]

Test:
Eliminate all favorites with morning line odds greated than 3/1
Result:
Eliminated 61 Horses in 61 Races
15 Eliminations per 100 finished at least 1st (85% accuracy)
36 Eliminations per 100 finished at least 2nd (64% accuracy)
52 Eliminations per 100 finished at least 3rd (48% accuracy)
64 Eliminations per 100 finished at least 4th (36% accuracy)

[RonTiller]

This is not encompassed by your database but one of the finest, if not THE finest throwouts in all of racing is:

Mountaineer morning odds 30/1 - all races since jan 1, 2005: 2 wins out of 942. Those 2 winners went off at odds of 43.1 and 61.2. No way to slice or dice this one - throw em out. Yee Haw!

Ron Tiller
HDW

[Tom]

I had 'em BOTH! :rolleyes::liar:

[podonne]

How about eliminate all horses in the race that are not one of the first four favorites in the wagering when the horses leave the paddock.

Well, I don't have the odds when the horses leave the paddock, but here's it based on the final odds:

3222 Horses eliminated in 854 races
0.17, 0.43, 0.78, 1.24 (83%, 57%, 22%, 0%)

[spilparc]

Test:
Eliminate all favorites with morning line odds greated than 3/1
Result:
Eliminated 61 Horses in 61 Races
15 Eliminations per 100 finished at least 1st (85% accuracy)


Sounds good to me. Just about what I thought. High priced morning line favorites win nowhere near 33%.

I'd like to see the same test done with a larger data base. I think the results would be about the same.

[podonne]

Just so everyone knows, I have the pps from every one of those races, so your eliminations can be more detailed than just odds.

Examples:

Need-to-lead horses who did well in thier last one or two races, but won't get the lead today. (Quinn, 2003)
Following a claim, a layoff of longer than 30 days followed by a drop in class below the claiming price (Quinn, 2003)
The last running line shows a win, but the horse failed to return for six weeks, except graded stakes horses who may be shopping. (Quinn, 2003)

[kenwoodallpromos]

Eliminate all dirt horses who have had more that 1 layoff of 60 days or longer showing on the PP's. (Turfers may lay off for the wintewr anyway).

[podonne]

Except for the occasional sprinter who might take the lead, horses four and older competing at new distances are best eliminated (Quinn 2003)

367 horses in 196 races
0.23 0.44 0.64 0.87

[podonne]

Eliminate all dirt horses who have had more that 1 layoff of 60 days or longer showing on the PP's.

Expression: ppcount("pp('dayssincelastrace') > 60") > 1
743 horses in 424 races
0.25 0.48 0.71 0.95

[douglasw32]

"0.25 0.48 0.71 0.95" what does it all mean ?

[podonne]

"0.25 0.48 0.71 0.95" what does it all mean ?
Ahh, sorry, chance a horse eliminated by the test will come in 1st, 2nd, 3rd, and 4th respectively.

Smaller the better.

While we're on the topic, I'd consider a good test to score

110 horses in 100 races
0.01 0.03 0.07 0.10

[douglasw32]

Thanks.

[Jake]

Well, I don't have the odds when the horses leave the paddock, but here's it based on the final odds:

3222 Horses eliminated in 854 races
0.17, 0.43, 0.78, 1.24 (83%, 57%, 22%, 0%)

I'm a little confused here. Are you saying a horse not in the top 4 final odds has a 17% chance of winning? Or, that 83% of winners will be in the top 4?

[Overlay]

As I interpret it, that's what he's saying. 83% of the time (five races out of every six), the winner of the race will come from among the four horses with the lowest post-time odds (the favorite, second choice, third choice, and fourth choice), and 17% of the time, the winner will come from outside that group of four. William Scott based his handicapping approach on something similar, using the statistic that two-thirds of all winners come from the three top betting choices. He then handicapped with the goal of predicting which one of those three would beat the other two in two races out of every three.

Scott also found that the winning percentages of betting choices got progressively smaller as the choice got higher. (In other words, favorites won on average 33% of the time, second choices 20%, third choices 14%, and so on in consecutive decreasing order.) Using that model, I would have expected the combined winning percentage for the top four horses in a race to be slightly lower that 83% (maybe something on the order of 77%), but perhaps the difference is attributable to smaller overall field sizes today compared to when Scott was writing, which would raise the percentage for each choice.

[prank]

I'm very curious to see how various people\methods go about "eliminating the non-contenders". Every handicapping book i read says this is the first thing to do, but I can't find any good statistics to back up the "easy eliminations" they recommend.

So I built this little number to try out a few ideas I have, but I don't have that many. Anyone have some ideas they've always wanted to test? I'll be glad to share the results here of every idea, or send to you directly if you prefer.


That's a great question. I'm very interested in something like this - basically iteratively reordering the ranking. (An aside, my area is statistics and an area called machine learning, and will confess to being the forum's low man on the totem pole with regard to handicapping.)

From the statistics literature, I don't believe there are many good methodologies that have been published. However, if you or anyone else would like to talk about trying some ideas, please PM me. If you're interested in longer-term research on this, I'd like to keep in touch, since I'm collecting data and experiments on how well different methods work.

I'd be very interested in knowing what the handicapping literature says about eliminating the non-contenders.

Thanks!

pRank

[hurrikane]

Ahh, sorry, chance a horse eliminated by the test will come in 1st, 2nd, 3rd, and 4th respectively.

Smaller the better.

While we're on the topic, I'd consider a good test to score

110 horses in 100 races
0.01 0.03 0.07 0.10

I'm not so sure this is the only thing to consider. layoff horses win less than a non layoff horses but are overlays and can show a pos ROI (under certain circumstances).

you suggest eliminating horses 8-1 or higher but at small class tracks with terrible lines makers, MSW, MCL, 3yo, turf races this is a big mistake. I, in fact, look for horses outside this parameter with certain positives going for them.

My point is, because the data may support a theory it is not telling the whole story. You have to dig a little deeper than your numbers.

Just my opionion but an interesting thread. My elimination factors concern numbers unrelated to numbers most up here would have so I can't really offer much here on methodology. All I can offer is that I would be hesitant to eliminate any horse based on some hard fast rule. You might at least give em a look through first.

[Overlay]

All I can offer is that I would be hesitant to eliminate any horse based on some hard fast rule. You might at least give em a look through first.

I've also found it to be more productive to try to assess the chances of all horses in a race to be sure I'm not missing a possible longshot winner, rather than eliminating one or more horses outright without even looking at their records for positive signs, or for parimutuel value. To me, any extra time involved is compensated by the prospects of a higher return than would come from narrowing a field down to a final choice that generally ends up at low/underlaid odds because everyone else is using the same elimination approach. (That's not to say that I would bet blindly or exclusively on high-odds horses, but it alerts me to the presence of possible winners that the public may not have taken sufficient notice of, and that are worth a bet in light of the odds at which they're being held.)

[twindouble]

Except for the occasional sprinter who might take the lead, horses four and older competing at new distances are best eliminated (Quinn 2003)

367 horses in 196 races
0.23 0.44 0.64 0.87

Never fails to blow my mind when I read things like this, many 4yo's are just coming into their own, to say they can't handle a change in distance is way off base. Also there's a wide variety of conditions were older horses can compete going longer or shorter. When anyone applies rules like that it's like saying you can only use hamburg for meat loaf. I cook.:D

Many times I've caught good breaking older routers shorting up against closing sprinters including hard knocking 6f horses that set a good pace ESP at middle distances 6 1/2 to a mile. Those same horses have the stamina to handle off tracks as well. Take a look at Trickey Trevor, 5th at Churchill yesterday and tell me the change in distances hurt that horse, I jumped all over that one, win and in the gimmicks. A 7 year old. You have to look at the race in hand, not stats that lock you into a rules. That's just one example, the majority of horses start out running sprints anyway.


T.D.

[ryesteve]

I like where this thread is trying to go, but you guys really need to look at ROIs, not just hit percentages.

[prank]

Never fails to blow my mind when I read things like this, many 4yo's are just coming into their own....


That reminds me of another topic that's missing: confidence in the estimates for a given horse's capabilities. Does anyone do that? From a talk given by Bill Benter, I can see that he seems to have looked at that. In some other contexts, such as chess rankings, it comes up (movement in Elo scores tend to tighten as one plays more games).

Twindouble's comment makes me think of a couple of things: when do you decide that there's not information on a horse to place a bet on it, and when do you decide there's not enough information for all the horses in the race to place any bets? I'll start a new thread on that topic, rather than distract from this one.

Thanks!

prank

[BetHorses!]

There was a Beyer Elimination system I used with much success many years ago. What you do is this.

Take the last 3 beyers from every horse and circle the lowest one for each. Now find the highest lowest number and subtract 4 points. Now go back and eliminate any horse who has not run better than this number in his last three

Example last 3 beyers I bolded lowest

#1 89 92 88

#2 85 88 79

#3 67 64 63

#4 71 70 72


88 is the highest low. Subtract 4 and 84 is your target. Now any horse whose highest beyer in his last 3 which is 84 or less is tossed. #3 and #4 is eliminated in this example. This worked very good but I have not tracked it for some years now. Believe it or not many short priced horses fit this profile and did not win. I would be interested if anyone uses this or has any data for it.

[Tom]

BH...I have that method and used to ue it back in the day. I thought it was a pretty good way to get contenders, but I never kept any data on it.

[Jake]

Ahh, sorry, chance a horse eliminated by the test will come in 1st, 2nd, 3rd, and 4th respectively.

Smaller the better.

While we're on the topic, I'd consider a good test to score

110 horses in 100 races
0.01 0.03 0.07 0.10

The reason I asked is that he indicated the horse itself--not the group of non-qualifiers in the race--had the given hit rate. So, when he tests, he also needs to calculate the number of horses in the race that match parameters, to adjust his actual percentages accordingly.

[podonne]

Whew, I should've checked more often, but let me clairify a few things.

The idea is to identify a certain test, say Test A. If I apply test A to race 1 and it eliminates horse A, what is the probability that Horse A will finish 1st? (or 2, or 3 ...)

There is no attempt here to identify winners, or rank horses.

Also, I dont profess any of the ideas here, most of them arent any good, Im giving the chance to test some common ideas that may not be as good as you might think. for example, Quirn's tests above aren' any good, but people qoute them all the time in books as sure things.

I'll write more when the plane lands.

[podonne]

Bet Horses, that’s good, I actually ran that on a limited 100 race sample on the plane. Failure rate for the win slot was only about 0.15. Not bad.

It was better if you increased the subtraction number past four. Ill run on a larger set tonight and let you know.

[bpiets]

:cool: From the drf 'i' generally look ar the 2 morning line fav.'s and do comparisions of the remaining field against 'them'....then eliminate all those that ''appear'' to be in over their heads...and then look for a ''spot'' play and if there isn't 'one' 'i' will go along with the favorite and ''those'' in exotics that ''appear'' to have some ''sort'' of advantages/possibilities that the others in the race don't have...generally from the past or a knowledge of how the 'horse' ran last week or in the recent past... :)

[Valuist]

I eliminate any horse who hasn't run a figure in the past year (at a similar distance over similar footing) that could run second in today's race, or about 3 pts within par. I play a lot of exactas and look for longshot runups, so if you are primarily a win bettor, you could make the qualifications more restrictive, maybe run par or better within the past 6 months.

I don't believe in blindly eliminating too many right off the bat. The DB people can quote all the numbers they want but some of the best value bets are the 4th, 5th and 6th choices in the betting, especially when the favorite is overbet. I think those DB numbers may be right but what doesn't tell the whole story is there's quite a few races where only 2 horses are legit win contenders, especially in small fields. And those type of races may distort things, leading to generalizations like "only focus on the top 4 betting interests".

[Jake]

How about eliminate all horses in the race that are not one of the first four favorites in the wagering when the horses leave the paddock.

Well, I don't have the odds when the horses leave the paddock, but here's it based on the final odds:

3222 Horses eliminated in 854 races
0.17, 0.43, 0.78, 1.24 (83%, 57%, 22%, 0%)
Hi, Podonne,

It appears that 17% represents the percentage of non-top 4 favorites that can win for a given race, with multiple per race eliminations. So, that the individual horse's win rate becomes roughly 6%. Is that 's how you are reading this? Great idea for a thread, thanks.

Jake

[BetHorses!]

Bet Horses, that’s good, I actually ran that on a limited 100 race sample on the plane. Failure rate for the win slot was only about 0.15. Not bad.

It was better if you increased the subtraction number past four. Ill run on a larger set tonight and let you know.



ok, thanks. Past 4 you do not get many of the shorter priced horses though.

Hope you are not the pilot :)

[betovernetcapper]

Like a lot of guys my first exposure to racing literature was Tom Ainslie. Tom had various rules for throwing out horses. One of the reason he gave for this was that throwing out non-contenders left more time for for creating pace ratings & looking at chart books. Given that all work had to be done by hand throwing out horses was a blessing.
Before Beyer numbers came out in the form, you had to make your own, horse by horse race by race. Whatever you could toss you didn't need to make a fig for.
In the early days of computer handicapping, you had to type in every stinking number and hope that you didn't make a mistake. Again throw outs came in handy.
With todays handicapping tools, I don't consider eliminatons very important.
I can press a button and see various pace matchups-anothor click and I see the bias. I'm going to be playing a race in a few hours off a trainer angle that took just another click.
Another factor in eliminations (for me) is that each track is different and winnig factors tend to be different. Automatic universial rules are unlikely to work.

[podonne]

Hi, so couple more things,

Jake and some others have pointed out that the probabilities are a bit schewed, and they are right, probably because I didn't explain it correctly. The actual mathmatical phrase is something like this:

"Given that Test A eliminated at least one horse in a given race, let Pa, Pb, Pc, and Pd be the probabilities that any of those horses eliminated will come in 1st, 2nd or better, third or better, and fourth or better respecitvly. "

That last probability Pd can also be thought of as the probability that a superfecta bet covering all of the other horses not eliminated by the test will fail. So you could say that a test has a 0.10 probability of failure, and if it eliminated two horses each has a 0.05 probability of failure (cumulativly 0.1), but you have to eliminate both together, so its a bit of a non-sequiter.

Hope that helped. I'm afraid I'm not explaining myself very well.

One more thing, I've seen a few comments denying that this analysis has any worth, and I agree that broad tests dont do any good. But I encourage those people to think a bit outside the box. Go to the extreme! Read BetHorses! post above for a good example. No its not particularly accurate as stated, but if you increase his subtraction figure to 20 instead of four, you could've eliminated 21 horses in 20 races in my 100 race sample (1 in 5!) at a 0.00% probability of failure.

Think crazy! What if you picked up a Derby form and there was a $5,000 claiming horse running in it? Could you eliminate him? What about a horse who ran once as a two year old and is making his second start as a 10 year old?

Yeah it won't happen often, but if you string enough tests together you might hit on that one race a day where enough of those crazy ideas lined up to eliminate the field down to a managable number, and BAM!

[podonne]

Also, a perfect example of accepted wisdom gone wrong, which I encourage you to put your nuggets to the test.

Andy Beyer: "The surest bet in racing. If the youngest horse in the field is running with the highest weight, he will always lose." - Speed to Spare

Happened 4 times in my sample, horse won twice and showed a third time. Not quite as sure a bet as Andy thought.

[spilparc]

here was a Beyer Elimination system I used with much success many years ago. What you do is this.

Take the last 3 beyers from every horse and circle the lowest one for each. Now find the highest lowest number and subtract 4 points. Now go back and eliminate any horse who has not run better than this number in his last three

Example last 3 beyers I bolded lowest

#1 89 92 88

#2 85 88 79

#3 67 64 63

#4 71 70 72


88 is the highest low. Subtract 4 and 84 is your target. Now any horse whose highest beyer in his last 3 which is 84 or less is tossed. #3 and #4 is eliminated in this example. This worked very good but I have not tracked it for some years now. Believe it or not many short priced horses fit this profile and did not win. I would be interested if anyone uses this or has any data for it.

Read BetHorses! post above for a good example. No its not particularly accurate as stated, but if you increase his subtraction figure to 20 instead of four, you could've eliminated 21 horses in 20 races in my 100 race sample (1 in 5!) at a 0.00% probability of failure.

I'm not quite following you here. Are you saying you could have eliminated one horse per race--approximately?

If you subtract 20 you won't be eliminating anybody, will you? Won't everyone qualify, with an occasional rare exception?

(I am curious where the original four came from, however.)

Thanks

[No Boy]

Eliminate any horse who did not finish 4th or better in either of its last two starts. Also accept as a contender horses who finished 5th last out when they have only one lifetime start.

[podonne]

If you subtract 20 you won't be eliminating anybody, will you? Won't everyone qualify, with an occasional rare exception?


Spilparc. YES! absolutely! In fact, that's the idea. And hopefully when that rare exception does occur, you can eliminate him with absolute confidence.

Also, Technically its that there were 20 races where you could've eliminated at least one horse, and in those 20 races, 21 horses total were eliminated.

[ryesteve]

Spilparc. YES! absolutely! In fact, that's the idea. And hopefully when that rare exception does occur, you can eliminate him with absolute confidence.
And then what? Even if this rule was actually infallible (which of course it isn't), what would generally happen is that you'd manage to eliminate a 70-1 shot from a 9 horse field. I guess you can argue "it's a start", but I don't see how it really changes how you'd look at the race.

[podonne]

And then what? Even if this rule was actually infallible (which of course it isn't), what would generally happen is that you'd manage to eliminate a 70-1 shot from a 9 horse field. I guess you can argue "it's a start", but I don't see how it really changes how you'd look at the race.
It is a start, and that particular test may not change the way you look at the race, but again:

1) My purpose in this thread was to give everyone a chance to empiricaly test eliminations they think are good.

and

2) If you have enough of these tests and put them on enough races, well that might just change the way you look at a race.

[bpiets]

:cool: Another ''way'' to bet is to 'log-on' in as many forums that have a contest and check the 'con-sences' area per race and compare 'your' selection to those of other bettors....the idea here ofcourse is to 'know' who the better-bettors are and ''reflect''/''compare'' 'your' selection whith 'his/hers'....this may be a bit difficult at times but a 'con-sence-is' can be most helpfull in the long run...and then there's the 'spread-sheets' that do comparision anali-sis....

[podonne]

Eliminate any horse who did not finish 4th or better in either of its last two starts.

(I added that he has to have at least two starts)


As you originally put it:


1217 587 0.14 0.30 0.50 0.78 (never better than 4 , past 2 starts)


It gets more interesting if you increase the four or increase the number of starts it has to look back through


367 265 0.07 0.15 0.23 0.39 (never better than 6 , past 2 starts)
472 289 0.07 0.18 0.30 0.49 (never better than 4 , past 5 starts)



An interesting note. It gets more accurate as you increase the "never finished better than #" parameter, but it actually appears to get less accurate as you look past five starts



369 216 0.08 0.21 0.33 0.55 (never better than 4 , past 8 starts)

[podonne]

I eliminate any horse who hasn't run a figure in the past year (at a similar distance over similar footing) that could run second in today's race, or about 3 pts within par.

I added a few things: Had to have at least five races in the past year, "similar distance" was within one furlong (220 yards).

Didn't turn out so well:
2786 693 0.52 1.03 1.62 2.17 (within 3 points of par)

even at the extreme (which is a bit suprising):
229 169 0.15 0.27 0.41 0.57 (within 51 points of par)

Feel free to revise.

[podonne]

There was a Beyer Elimination system I used with much success many years ago. What you do is this.

On further testing, not as good as I thought with that small sample. Your original parameters tested about the same:

560 331 0.14 0.27 0.46 0.70 (subtract 4, last 3 races)

It was most accurate at 10,3

151 118 0.04 0.08 0.24 0.38 (subtract 10, last 3 races)

but tended to get less accurate as you subtracted more than 10

33 28 0.14 0.21 0.46 0.61 (subtract 20, last 3 races)

[douglasw32]

How about that test on finish position using...

Has not beat 1/2 the field in last three races ????

[spilparc]

On further testing, not as good as I thought with that small sample. Your original parameters tested about the same:

560 331 0.14 0.27 0.46 0.70 (subtract 4, last 3 races)



Using the 4 subtraction method what happens if you eliminate races where there are two or more first time starters and/or two or more foreign horses with no American lines and/or two or more horses who have never run on the surface?

Also, don't use dirt speed ratings in turf races and don't use turf speed ratings in dirt races.

[ryesteve]

1) My purpose in this thread was to give everyone a chance to empiricaly test eliminations they think are good.
As I said in an earlier post, it's a worthy goal, but with no one actually talking about the monetary performance of the horses that are targeted via these proposed rules, I don't see what's being learned. For instance, if it turns out this Beyer rule, over a large sample, identifies 98% losers, does it really help if it turns out the average winning mutual is $90?

[podonne]

How about that test on finish position using...

Has not beat 1/2 the field in last three races ????

Brisnet.com PP drf files seem to have only "Number of Entrants", which may or may not be adjusted for how many horses actually start in the race. Hopefully it does:

828 461 0.15 0.30 0.48 0.73 (has not finished better than 50% of the field, in last 3)

Again, it gets better as you get more restrictive

113 103 0.04 0.10 0.17 0.31 (has not finished better than 80% of the field, in last 3)

and here was the best result:

43 41 0.00 0.02 0.07 0.24 (has not finished better than 80% of the field, in last 4)

[douglasw32]

COOL, not sure what it all means, but I agree with what your doing if nthing else the potential to STOP DOING SOMETHING SILLY because it has been repeated over and over as being a good thing...

[skate]

i think that so.cal. has the right idea.

if you can eliminate the first 3 or 4 (lowest odds), then you have a playable race.

[Jake]

I added a few things: Had to have at least five races in the past year, "similar distance" was within one furlong (220 yards).

Didn't turn out so well:
2786 693 0.52 1.03 1.62 2.17 (within 3 points of par)

even at the extreme (which is a bit suprising):
229 169 0.15 0.27 0.41 0.57 (within 51 points of par)

Feel free to revise.

Some clarification, please. Is the horse within 3 points of par above, or falling short by more than 3 points of par? I assume Valuist is eliminating any horse not within 3 points of par.

[Jake]

As I said in an earlier post, it's a worthy goal, but with no one actually talking about the monetary performance of the horses that are targeted via these proposed rules, I don't see what's being learned. For instance, if it turns out this Beyer rule, over a large sample, identifies 98% losers, does it really help if it turns out the average winning mutual is $90?

Isn't this a similiar approach to a negative UDM in Jcapper? Once you find a poor win percentage, wouldn't you follow up and check for ROI exceptions? If you can identify that $90 exception with other factors, fine, but otherwise it is preferable to toss him along with the other 98% losers. I have been using a rule based system for many years, and the approach that's proven best is top-down: generalize and then find exceptions (sub-rules). By extension, this is also a robust approach for deriving accurate odd-lines. Of course, I also believe that high ROI spot play UDMs are a great approach as you have indicated; I just don't find the different approaches mutually exclusive.

[ryesteve]

Isn't this a similiar approach to a negative UDM in Jcapper? Once you find a poor win percentage, wouldn't you follow up and check for ROI exceptions?
Honestly, no. If I'm trying to come up with a negative UDM, I'm not looking at the win%... I'm just looking at the ROI. If my proposed criteria doesn't have a poor ROI, then I don't consider it to be "negative".


If you can identify that $90 exception with other factors, fine, but otherwise it is preferable to toss him along with the other 98% losers.
Again, I don't see the value in tossing out horses whose ROI is the same (or in this case, better) than the ROI of the horses you keep.

[Jake]

Honestly, no. If I'm trying to come up with a negative UDM, I'm not looking at the win%... I'm just looking at the ROI. If my proposed criteria doesn't have a poor ROI, then I don't consider it to be "negative".


Again, I don't see the value in tossing out horses whose ROI is the same (or in this case, better) than the ROI of the horses you keep.

Thanks for the reply. My working assumption is that a positive ROI horse is a function of estimated win % relative to actual odds. So, if you have a system that consists of only two rules to determine a horse estimated win %, would you say it has no value?

Given this thread, let’s say the first rule is that horses in the top 4 favorite slots will win 83% of time, the remaining horses only 6% (assuming average of 7 horse fields). The second rule states that of the remaining horses, those 6%ers, only the gray horses at 10-1 or above win. The rest always lose. In other words, you can now compute a conditional probability that improves your prior estimate win % for all the horses. Since estimate win % is one of the key drivers for ROI, the value here is implicit

[ryesteve]

Given this thread, let’s say the first rule is that horses in the top 4 favorite slots will win 83% of time, the remaining horses only 6% (assuming average of 7 horse fields). The second rule states that of the remaining horses, those 6%ers, only the gray horses at 10-1 or above win. The rest always lose.
The way I'd interpret this is that it's the 2nd factor that has the wagering value, not necessarily the first. I get the gist that you're looking for an interaction effect, but there's no reason to start looking for interactions of multiple factors when there are self-contained elimination rules that could be applied first.

[Jake]

The way I'd interpret this is that it's the 2nd factor that has the wagering value, not necessarily the first.

No, they both have wagering value, that's what makes it conditional. The 2nd factor allows you to adjust your prior win probablities, identify your contender group, and eliminate non-contenders. ROI value is derived from those revised win probablities against actual odds. I'm not suggest anyone handicaps this way, rather just trying to explain how Podonne's examples, which are not transparently ROI-based, have wagering value.

[ryesteve]

No, they both have wagering value, that's what makes it conditional.
But until you show the gray horses at 10-1 don't outperform within the 1st second, you don't know that it's conditional.

And so far, I don't think I've seen anyone here attempt to link any factors to investigate interaction effects. All we're seeing is one rule at a time, which may or may not be identifying horses that underperform in terms of ROI. So if all we're looking at, at this point, is a single rule at a time, the ROI shouldn't be ignored.

[podonne]

Jake,

I suppose it should've read "not within 3 of par". I did respect his idea in the test.

Also, I'm a bit perturbed by the way the conversation always returns to an ROI discussion. In the first stage of handicapping when you are eliminating the non-contenders, why are you concerned with the ROI of the horses you eliminate? Unless you don't think you can eliminate any contenders blindly (which I think I've shown is possible) and you are using the tests to rank horses, which is not the intention of this research and its actually dangerous to use this research in that way...

-Philip

[formula_2002]

Jake,

Also, I'm a bit perturbed by the way the conversation always returns to an ROI discussion. In the first stage of handicapping when you are eliminating the non-contenders, why are you concerned with the ROI of the horses you eliminate?

-Philip

The roi comparison between your "method" and what the public would normally acheive for horse's in the same odds range, will allow you to begin to gauge if you have found something the public is not aware of.

Another way of looking at it.
If the non contenders at wining 6 in 100 races, and their average odds is say 2-1, great.
But if their odds are about 16-1, well, you and the public are onto the same thing..and that aint good ;)

[ryesteve]

Also, I'm a bit perturbed by the way the conversation always returns to an ROI discussion. In the first stage of handicapping when you are eliminating the non-contenders, why are you concerned with the ROI of the horses you eliminate?
And I'm equally perturbed by those who insisting on ignoring it. Why would someone NOT be "concerned with the ROI"? The idea is to make money, not select winners. I can bet every horse in the race and lose 20% of my money. If I'm able to eliminate half the horses in a race via this kind of analysis, and yet I'm still losing 20% of my money on the so-called "contenders", I've accomplished nothing. There's no value in seperating "contenders" and "non-contenders" into two groups that show the same monetary performance.
On the other hand, if the half of the field I eliminate loses 30 cents on the dollar, then my contenders are only losing 10 cents on the dollar, which means I'm that much closer to showing a profit.

[podonne]

Yes, but, the idea here is not to bet the non-contenders, in fact, the idea is to show which horses you should not bet on in any case because they will lose.

[podonne]

Ahh, ok wait. You guys are talking about using a particular test as a seperator and determinine which set of horses provides the best ROI. That's not my position. I want to identify many tests that each eliminate 1 horse in 20 for many different reasons, but are really sure when they do make an elimination that that horse will not win.

If you want to talk about ROI in my context, you're talking about expected payout models and I'm not nearly that far yet.

[ryesteve]

Yes, but, the idea here is not to bet the non-contenders, in fact, the idea is to show which horses you should not bet on in any case because they will lose.
Yes, we understand no one wants to bet the NON-contenders. The point is, if the non-contenders UNDER-perform (monetarily) then it follows that the contenders OVER-perform (monetarily). Finding those horses who over-perform (monetarily) by a wide enough margin is the only way to show a profit.



the idea is to show which horses you should not bet on in any case because they will lose.
There is no rule that will select 100% losers; and if there were, then the ROI is -100%, in which case I'd say you had a great rule. But in real-world terms, maybe you'll come up with a rule that picks 95% losers... but just knowing THAT figure tells me nothing. As Joe M. was saying, if you're able to select a group of horses that loses 95% of the time, without knowing what the odds on those horses were, you could have anything from an awesome elimination tool to an awesome longshot selection tool.

[Jake]

But until you show the gray horses at 10-1 don't outperform within the 1st second, you don't know that it's conditional.

Sorry, I have no idea what you are saying here.


Let's put this to bed. If you have a accurate win estimate, you can always determine if a horse is or isn't a overlay in a given race. Whatever goes into your method to give you that win estimate has handicapping value. You can use the queries here to determine a win estimate.

ROI numbers has tremendous handicapping value as well, but it does not negate the win% value that Podonne has posted in his queries. Like you, I prefer to play high ROI price horses. I find them using both win% and ROI research in combination. ROI testing without the base win rate is extremely misleading--but that's another discussion.

So, in honesty, I appreciate the points you have made. But for personal peace of mind, if you find what Podonne is doing here of no value or the methodolgy short of your requirements, simply quit reading the thread. As Podonne has posted above, his purposes for the thread is apparently tangential to our discussion in any case.

[ryesteve]

if you find what Podonne is doing here of no value or the methodolgy short of your requirements, simply quit reading the thread.
It's not as if this is a private conversation I'm butting into. There are plenty of folks reading this, and I don't see the harm in trying to make the point that if one is trying to eliminate horses, you need to eliminate those who lose a dispoportionate amount of money, not a disproportionate % of races. Even if the original poster disagrees, at least those following along are being given something else to think about.

[Tom]

http://www.netcapper.com/TrackTractsArchive/TT010126.htm


This is based all on roi......

[Jake]

It's not as if this is a private conversation I'm butting into. There are plenty of folks reading this, and I don't see the harm in trying to make the point that if one is trying to eliminate horses, you need to eliminate those who lose a dispoportionate amount of money, not a disproportionate % of races. Even if the original poster disagrees, at least those following along are being given something else to think about.

Fair enough. I would say that your point has been asserted enough times for everyone to get it. The same with my point of view. Again, I suggest this all be put to bed. Otherwise, if you are so inclined, just post the same queries with the ROI values from your own program data, and everyone will be happy.

With respect, that's all I'm going to say on this subject.

[ryesteve]

http://www.netcapper.com/TrackTractsArchive/TT010126.htm

This is based all on roi......
Good stuff there!

[formula_2002]

http://www.netcapper.com/TrackTractsArchive/TT010126.htm


This is based all on roi......

Good post,
I checked out the ml 30>=1
found 4255 horses that returned an roi of .38.
Went a bit further.

the average odds of those horse was 67-1
that represents 1.47% probability.

At a 17% take track the book (sum 1/(odds+1)), is about 1.204.
If you could deduct the full 1.47 from the book,
that would make the book (1.204-.0147)=1.1893, lowering the take to 16%..
But I think you can deduct the FULL 1.47% only if the horses lost 100% of the time!!.

In this case, they won 1.12 % of the time, so I would think that you DONT get the full 1.47% advantage, rather its .0147*(1-.0112) = 1.45%.

That may not sound like much, but the concept is important, especially when you think you can break even by eliminating non contenders who's sum (1/(odds+1)=the track take..Not going to happen, unless the non condenders lost 100% of their races.

At a track with a 15% take-out, eliminating non contenders who's sum(1/(odds+1)=.15, and lose all the time, the vig is still 1.176-.15=2.6%..
It's good to eliminate non contenders, I just want to estimate its value correctly.

Peer review always welcome!

Joe M

[podonne]

http://www.netcapper.com/TrackTractsArchive/TT010126.htm


This is based all on roi......

Wow that was a great post. And I like where the thread is heading to, and based on overwhelming demand, I'm going to add an ROI calculation to the software.

Turf race, horse has one of the top four last-race Beyer numbers, last race was on firm course, today is an off-track (.44 ROI).

Favorite today, has a layoff of 45+ days between the last and the second-to-last race (.57 ROI).

I'm still interested in finding eliminations like this to test (although I only have fast dirt races), especially tests which don't rely on odds to make the eliminations, but things like class, form, speed, etc...

[skate]

you can (oh, you can if you want to) NOT eliminate a horse on Speed, Pace, Class, even Form from a race.

but "odds" alone can do this very thing, eliminate the horse if odds are too low, hey that's me.

1000/1 may by itself be a good bet, maybe not too likely, but still good. gees, put him in the tri at least.
and that makes you a winner.

i agree, it is not winners i'm after, but my return is very somewhat significant. ha