Predicting Speed Ratings from Past Ratings

[classhandicapper]

This thread was eventually closed, but aspects of it were very interesting. I'm giving it a bump hoping that we can continue the conversation in a productive way partly because I think I know more now than I did back then. I may contribute and get more out of it this time around....as soon as I reread what I was thinking back then. :lol:

http://www.paceadvantage.com/forum/showthread.php?t=118142&page=1&pp=15&highlight=trifecta+mike

[aaron]

This thread was eventually closed, but aspects of it were very interesting. I'm giving it a bump hoping that we can continue the conversation in a productive way partly because I think I know more now than I did back then. I may contribute and get more out of it this time around....as soon as I reread what I was thinking back then. :lol:

http://www.paceadvantage.com/forum/showthread.php?t=118142&page=1&pp=15&highlight=trifecta+mike
As for predicting speed ratings from past ratings,in my opinion there are many horses you can do this with,especially if you are familiar with the horses trips and his trainer. For example there are one number horses who always regress off the number. Then there are horses who run the big number off a positive trainer change and are a good bet to repeat the number in their next race. Also,there are horses who almost always run their number and if in the right race can be good bets.

[raybo]

Personally, I think that there was some interesting discussion in that thread. The fact that TM stated that the initial statistical analysis that indicated that the 3rd race back was the least significant of the last 4 race's speed figures, was where many people "got off", and that was unfortunate.

TM stated that he supplied the stats guy with a full year of race data, from all tracks. That was meaningful info, although it appeared to be ignored by many. It was meaningful, IMO anyway, in that any results obtained from that data, regarding individual races in the future, would not be of particular value. And also, any results from that data, only applies to that full year of data, from all tracks.

So many times posters here insist that someone provide a test of a principle that has been discussed, by posting a real race example (not to say that there have been many times where a poster has actually challenged someone to a contest, obviously meaning a very small sample of individual races). They invariably fail to recognize the fact that long term data, and its resulting long term implications, probably will only reveal itself, in the long term, not in smaller samples of individual races.

The "next race" mindset may be important to most players, but that same "next race" attitude will more than likely fail over the long term. In other words, some of those kinds of players will experience good results (or bad results, or somewhere in between), in the short term, but will still fail long term, exactly due of that "next race" focus.

In short, small samples are just small samples, and will vary widely among all players, and all methods. Most will also fail in the long term, but a very few will be successful in the long term. This long term failure is a fact of racing life for approximately 98% of all players.

A year's worth of data, from all tracks, is of little value to anyone not playing for the long term (more than a year), at all tracks. This will be true for almost every player in the game. For the data to be of use, one must use those results/implications as only one of the tools one uses in playing the races, and one would probably be better served by not focusing solely on the "next race".

Really, when you look at any number of factors, or combinations of factors, like speed figures, the predictive-ness will vary from race to race, day to day, week to week, etc., just like anything else in racing. So, if you know that it will vary, why focus on a single race, or a single day, or a single week, etc.?

Regarding TM's OP and his stats guys' findings, those findings were based on what TM wanted to know, that being how to apply values to the weightings of the last 4 speed figures. I think that one could very well arrive at those values, as was presented by several in that thread, albeit from differing methodologies. However, the underlying point that the 3rd figure back was the least important of the last 4 figures, has little significance to me, only that including the 3rd back figure would not add significance to the results. So what, it also does not detract from the overall results significantly. So, use all 4, with the weightings you received from the analysis, add up all 4 resulting calculations for each horse, obtain your projection for all horses in today's race, and then perform your other "due diligence" and make your determinations for betting. I never thought that TM's purpose was to use only this speed figure methodology in isolation, but only as one of several methodologies to arrive at betting selections. Probabilities are just probabilities, there is still much other work to do in order to produce long term profit.

[Tom]

Today - 6 furlongs, dirt, fast

Horse A
6f dirt sly 90
5f turf frm 66
6f dirt fst 95
6f dirt fst 93

Horse B
6f dirt fst 71
6f dirt fst 83
6f dirt fst 82
6f dirt fst 77

Do you want to weigh both these horses the same?

[raybo]

Today - 6 furlongs, dirt, fast

Horse A
6f dirt sly 90
5f turf frm 66
6f dirt fst 95
6f dirt fst 93

Horse B
6f dirt fst 71
6f dirt fst 83
6f dirt fst 82
6f dirt fst 77

Do you want to weigh both these horses the same?

I'm not trying to be argumentative or inflammatory but, you just made my point Tom. Your example is focusing on a single example, the "next race" attitude I was referring to in my post, when the data used in TM's thread was for a full year, at all tracks. In short, the weightings may not apply to your single example. But, those weightings could very well apply to a year's races (or more), and if you play every track, every day.

[thaskalos]

So many times posters here insist that someone provide a test of a principle that has been discussed, by posting a real race example (not to say that there have been many times where a poster has actually challenged someone to a contest, obviously meaning a very small sample of individual races). They invariably fail to recognize the fact that long term data, and its resulting long term implications, probably will only reveal itself, in the long term, not in smaller samples of individual races.



When we ask for a "real-life example"...we are not asking so we can assess the idea's "long-term" worth. We know that a single isolated example isn't enough for that. We are also not asking for any "secrets", whose sharing could hurt in any way the person who presents this "real-life example". We ask for an actual example so we can see if the idea can actually be "applied" to the handicapping process...or if it's just a theoretical soundbite. In that perspective...just one example is enough...IMO.

Let's say that you came to this board and announced that you have discovered a brand new way of combining speed and pace figures when handicapping a race. I, or Tom, may ask you for an "actual example" of what you are talking about. By asking...we are not trying to determine if your new idea is "long-term profitable". We are also not asking you to share the "mechanics" of this idea with us. All we are asking you to do is APPLY these new numbers of yours to an actual race yet to be run...so we could see if this new idea actually leads to an actual handicapping CONCLUSION...REGARDLESS of how "accurate" this conclusion turns out to be.

All of us know that there is a difference between "theory" and "practice". We all enjoy handicapping "theory"...because, after all, this is our favorite game. But we like it even more if this "theory" could actually be applied to the handicapping task at hand. That's why we ask.

[Tom]

Exactly.
Is the point of all this that this race, while fairly obvious as to the answer, should be relegated to the pattern of thousands of other races and today handicapped incorrectly thinking you will make up this loss with other winners?

[classhandicapper]

The reason I resurrected this thread is because I've done some more studies since then. We all know that occasionally a horse's last race has to be dismissed or discounted, but the thing I am finding is that it's fairly difficult to outperform a horse's last race alone by trying to combine multiple races in a systematic way.

I even tried some approaches I intuitively thought should work, but didn't.

For example, I looked at a horse's last race, best of last 3 and best of last 5.

I assumed that if I combined best of last 3 or best of last 5 (which performed reasonably well on their own) with last race and weighted it so that they were only strong enough to serve as a tiebreaker when 2 horses were otherwise very similar off their last race, it should outperform last race alone. But it did not. I tried quite a few things with various weights and nothing systematic worked. The results would be similar, but last race alone would eek out a slightly better result.

I'm starting to think you need a very good reason to look much past a horse's last race. Combining races may find winners that last race won't get, but they will just be different horses. There won't necessarily be more of them or better values.

[thaskalos]

The reason I resurrected this thread is because I've done some more studies since then. We all know that occasionally a horse's last race has to be dismissed or discounted, but the thing I am finding is that it's fairly difficult to outperform a horse's last race alone by trying to combine multiple races in a systematic way.

I even tried some approaches I intuitively thought should work, but didn't.

For example, I looked at a horse's last race, best of last 3 and best of last 5.

I assumed that if I combined best of last 3 or best of last 5 (which performed reasonably well on their own) with last race and weighted it so that they were only strong enough to serve as a tiebreaker when 2 horses were otherwise very similar off their last race, it should outperform last race alone. But it did not. I tried quite a few things with various weights and nothing systematic worked. The results would be similar, but last race alone would eek out a slightly better result.

I'm starting to think you need a very good reason to look much past a horse's last race. Combining races may find winners that last race won't get, but they will just be different horses. There won't necessarily be more of them or better values.
You say "combine multiple races in a systematic way". This is what TrifectaMike was proposing too, I believe...and that's why asking for an "actual example" is appropriate in cases such as this.

These races are "unique events"...which are governed by unique race DYNAMICS. Each horse's past-performance record is comprised of unique races, whose differing circumstances distinguish them from one another. To cavalierly label races as Race #1, Race #2, Race #3, and race#4...and to make broad determinations simply by the ORDER in which these races are listed on the page, is to miss the point ENTIRELY...IMO.

[Tom]

Is the point of this to look at 8 horses in a race and predict who will run the
highest figure today?

[classhandicapper]

You say "combine multiple races in a systematic way". This is what TrifectaMike was proposing too, I believe...and that's why asking for an "actual example" is appropriate in cases such as this.

These races are "unique events"...which are governed by unique race DYNAMICS. Each horse's past-performance record is comprised of unique races, whose differing circumstances distinguish them from one another. To cavalierly label races as Race #1, Race #2, Race #3, and race#4...and to make broad determinations simply by the ORDER in which these races are listed on the page, is to miss the point ENTIRELY...IMO.

I get what you are saying. I was arguing the same thing in the original thread. I use a more interpretive approach in my actual betting. But in a more general sense, I thought for sure there would be something to gain by looking at multiple races. I'm not seeing much evidence of that in my tests so far.

[classhandicapper]

Is the point of this to look at 8 horses in a race and predict who will run the
highest figure today?

My original research was done by someone else using my data. The goal was to predict the horse's next Beyer figure based on his previous 3 Beyer figures. In this latest test, I am using my own figures and trying to maximize both the win% and ROI for the top rated horse.

[Tom]

So you want a formula to use for all horses?
You would have to break it down by age, sex, surface......three year olds often have dramatic improvement that is recognizable by patterns that do not exist in older horses, and they don't typically show them on turf or poly, for example.

I always look to see if I can predict what each horse will run next our, but never use any kind of formulas or whatnot, just I know 15-30 Beyer points improvement is to be expected relatively frequently with three old males, and some proven signs that will be a tell for a move ahead. But I don't look at older horse that way at all.

[raybo]

When we ask for a "real-life example"...we are not asking so we can assess the idea's "long-term" worth. We know that a single isolated example isn't enough for that. We are also not asking for any "secrets", whose sharing could hurt in any way the person who presents this "real-life example". We ask for an actual example so we can see if the idea can actually be "applied" to the handicapping process...or if it's just a theoretical soundbite. In that perspective...just one example is enough...IMO.

Let's say that you came to this board and announced that you have discovered a brand new way of combining speed and pace figures when handicapping a race. I, or Tom, may ask you for an "actual example" of what you are talking about. By asking...we are not trying to determine if your new idea is "long-term profitable". We are also not asking you to share the "mechanics" of this idea with us. All we are asking you to do is APPLY these new numbers of yours to an actual race yet to be run...so we could see if this new idea actually leads to an actual handicapping CONCLUSION...REGARDLESS of how "accurate" this conclusion turns out to be.

All of us know that there is a difference between "theory" and "practice". We all enjoy handicapping "theory"...because, after all, this is our favorite game. But we like it even more if this "theory" could actually be applied to the handicapping task at hand. That's why we ask.

I think you could answer your own question by applying your own weightings to the last 4 races of each horse, then running many races through your database (if you have that ability) to see what correlations are returned from that set of weightings compared to the actual winners, then try another set of weightings, and repeat the process. How the weightings are determined is the problem. I suspect that was what TM was getting at. His "assistant" suggested that the 3rd race back was insignificant related to the other 3. While others suggested that the 4th should be weighted lower than the 3rd, so the 3rd would have more significance than the 4th. Who is correct?

This does not suggest that weighting the last 4 speed figures, multiplying the figures by their weightings, and then summing or averaging the resulting calculations would be good or bad. It does suggest that if you want to arrive at weightings for various factors, long term data may offer some insights that you might not normally expect.

The players that are not interested in doing research of this sort might ridicule it as being absurd, without knowing if it is absurd or not. The seemingly absurd just might not be so absurd after all, but you'll never know unless you do the research and test it going forward.

Certainly the example Tom posted would result in Horse A being the better horse, and running the better speed figure, if you adhere to any of the posted suggestions that the last race is worth more than the 2nd, and the 2nd race is more important than the 3rd, and either the 3rd is more important than the 4th or vice versa. Now, does that mean anything? No, because regardless of how you weight the 4 speed figures for both horses, horse A will always come out with a higher score, because 3 of 4 of horse A's figures were significantly higher than any of horse B's figures. But, over the long run, and under more normal examples, one might find that there is a correlation between weighting the last 4 speed figures and predicting a future figure, or at least in ranking the resulting figures.

My own weightings produced Horse A with a rating (not a speed figure) of 86, while Horse B produced a rating of 75.1.

[thaskalos]

I think you could answer your own question by applying your own weightings to the last 4 races of each horse, then running many races through your database (if you have that ability) to see what correlations are returned from that set of weightings compared to the actual winners, then try another set of weightings, and repeat the process. How the weightings are determined is the problem. I suspect that was what TM was getting at. His "assistant" suggested that the 3rd race back was insignificant related to the other 3. While others suggested that the 4th should be weighted lower than the 3rd, so the 3rd would have more significance than the 4th. Who is correct?

This does not suggest that weighting the last 4 speed figures, multiplying the figures by their weightings, and then summing or averaging the resulting calculations would be good or bad. It does suggest that if you want to arrive at weightings for various factors, long term data may offer some insights that you might not normally expect.

The players that are not interested in doing research of this sort might ridicule it as being absurd, without knowing if it is absurd or not. The seemingly absurd just might not be so absurd after all, but you'll never know unless you do the research and test it going forward.

Certainly the example Tom posted would result in Horse A being the better horse, and running the better speed figure, if you adhere to any of the posted suggestions that the last race is worth more than the 2nd, and the 2nd race is more important than the 3rd, and either the 3rd is more important than the 4th or vice versa. Now, does that mean anything? No, because regardless of how you weight the 4 speed figures for both horses, horse A will always come out with a higher score, because all 4 of horse A's figures were higher than any of horse B's figures. But, over the long run, and under more normal examples, one might find that there is a correlation between weighting the last 4 speed figures and predicting a future figure, or at least in ranking the resulting figures.

You objected, in your prior post, to the idea of asking the original poster for a "real-life" example of what he means. Now...you say that "we could answer our own questions, by doing our own research". Does this mean that a poster could start a thread here, introducing "theoretical concepts"...but we are "wrong" to ask him for a PRACTICAL example of what he is talking about?

If the original poster's intention is to just make an opening statement here and then disappear...then what's the point of starting the thread?

[raybo]

You objected, in your prior post, to the idea of asking the original poster for a "real-life" example of what he means. Now...you say that "we could answer our own questions, by doing our own research". Does this mean that a poster could start a thread here, introducing "theoretical concepts"...but we are "wrong" to ask him for a PRACTICAL example of what he is talking about?

If the original poster's intention is to just make an opening statement here and then disappear...then what's the point of starting the thread?

What I'm suggesting is that you could answer your own question, and if so, why ask it in the first place?

BTW, I edited my previous post for errors, and also provided my answer to the example Tom posted above.

[thaskalos]

What I'm suggesting is that you could answer your own question, and if so, why ask it in the first place?

BTW, I edited my previous post for errors, and also provided my answer to the example Tom posted above.

And I am suggesting that the original poster has the OBLIGATION to answer a few questions when he opts to start a thread here. I don't care WHAT his academic credentials are...he must be willing to clarify the point that he is trying to make. If all he is intending to do is dazzle his "followers" with his "esoteric brilliance"...then he should write his post...and send it to his friends here by personal message.

These threads are for PUBLIC view...and they invite INTERACTION. If the original poster doesn't like to answer a few questions...then he should consider starting his own BLOG.

[raybo]

And I am suggesting that the original poster has the OBLIGATION to answer a few questions when he opts to start a thread here. I don't care WHAT his academic credentials are...he must be willing to clarify the point that he is trying to make. If all he is intending to do is dazzle his "followers" with his "esoteric brilliance"...then he should write his post...and send it to his friends here by personal message.

These threads are for PUBLIC view...and they invite INTERACTION. If the original poster doesn't like to answer a few questions...then he should consider starting his own BLOG.

If you read TM's OP you will see that he is asking a question of other posters, not the reverse. He is not saying that he is right or knows the answer, he's just putting a thought out there and asking what others would say to it.

What question(s) would you like TM to answer, BTW?

[thaskalos]

If you read TM's OP you will see that he is asking a question of other posters, not the reverse. He is not saying that he is right or knows the answer, he's just putting a thought out there and asking what others would say to it.

What question(s) would you like TM to answer, BTW?

Ray...I'm not looking to start an argument with you. I've argued with you plenty in the past...and I haven't exactly enjoyed the experience. My argument isn't with you...it's with these "cryptic" posts, which certain posters submit without the inclination to offer any sort of explanation as a follow-up.

What good does it do if I tell you the question that I would like TM to answer...if I am positive that TM will never ANSWER it?

TM was "asking a question of the other posters"...you say. I disagree! TrifectaMike doesn't ASK legitimate questions; he asks questions in the same manner that SOCRATES was asking people questions, 2,500 years ago. These questions were tailor-made to show how little the OTHER guy knew about the particular topic at hand. But...once the other guy's ignorance was acknowledged...Socrates actually supplied some answers. TrifectaMike just sits there...probably secretly laughing at his audience's ignorance. And I just don't see the point of that.

Start a thread when you have something to share...I say. When you want to provide only an opening statement and then hide...then post this statement in someone else's thread.

[raybo]

Ray...I'm not looking to start an argument with you. I've argued with you plenty in the past...and I haven't exactly enjoyed the experience. My argument isn't with you...it's with these "cryptic" posts, which certain posters submit without the inclination to offer any sort of explanation as a follow-up.

What good does it do if I tell you the question that I would like TM to answer...if I am positive that PM will never ANSWER it?

TM was "asking a question of the other posters"...you say. I disagree! TrifectaMike doesn't ASK legitimate questions; he asks questions in the same manner that SOCRATES was asking people questions, 2,500 years ago. These questions were tailor-made to show how little the OTHER guy knew about the particular topic at hand. But...once the other guy's ignorance was acknowledged...Socrates actually supplied some answers. But TrifectaMike just sits there...probably secretly laughing at his audience's ignorance. And I just don't see the point of that.

Start a thread when you have something to share...I say. When you want to provide only an opening statement and then hide...then post this statement in someone else's thread.

I'm not looking to argue with you either, but I would like to know what question you would ask him about that topic. Obviously TM got his answer via Magister Ludi, but personally I don't know what to do with that answer, because I don't know what to think of this: Perhaps there is a cycle of equine performance consistency. W3 could be the minimum performance consistency point. Is he saying that the 3rd race back is generally the most insignificant speed figure of the last 4, or is he saying something else? And, is he saying that one can reliably ignore the speed figure (regarding long term play) for the 3rd race back, or is he saying something else? We all can point to exceptions to refute universally ignoring that 3rd race back, but how does it work out long term, regarding assigning weightings to the last 4 speed figures? Does it only apply to the sample of data he provided to his assistant, or does it apply to other sets of data?

He went on to state that he uses a "Hierarchical Bayes Model" to predict speed figures from past speed ratings, while he had previously stated that modeling was not involved in the exercise. So, what does that mean? Does he actually use the statistical study data, regarding the "insignificance" of the 3rd race back, or not, or does he use it for something else? My understanding of Bayes is that it offers data where lack of data is a problem. If that is so, then why worry about the significance of one speed rating over another? Wouldn't the use of Bayes take care of that on its own?

If I knew what "W3 could be the minimum performance consistency point" means maybe I could answer some of my own questions, without input from TM or Magister Ludi. Maybe a Google of that might explain that phrase more?

[thaskalos]

I'm not looking to argue with you either, but I would like to know what question you would ask him about that topic. Obviously TM got his answer via Magister Ludi, but personally I don't know what to do with that answer, because I don't know what to think of this: Is he saying that the 3rd race back is generally the most insignificant speed figure of the last 4, or is he saying something else? And, is he saying that one can reliably ignore the speed figure (regarding long term play) for the 3rd race back, or is he saying something else? We all can point to exceptions to refute universally ignoring that 3rd race back, but how does it work out long term, regarding assigning weightings to the last 4 speed figures? Does it only apply to the sample of data he provided to his assistant, or does it apply to other sets of data?

He went on to state that he uses a "Hierarchical Bayes Model" to predict speed figures from past speed ratings, while he had previously stated that modeling was not involved in the exercise. So, what does that mean? Does he actually use the statistical study data, regarding the "insignificance" of the 3rd race back, or not, or does he use it for something else? My understanding of Bayes is that it offers data where lack of data is a problem. If that is so, then why worry about the significance of one speed rating over another? Wouldn't the use of Bayes take care of that on its own?

If I knew what "W3 could be the minimum performance consistency point" means maybe I could answer some of my own questions, without input from TM or Magister Ludi. Maybe a Google of that might explain that phrase more?

That's my point...NO ONE knows what any of this means. And it was never INTENDED for any of us to eventually get to KNOW what any of this actually means.

TrifectaMike wanted to test our group here, in much the same way as a teacher occasionally chooses to test his class...to see which student is the 'brightest". He asks a cryptic question of the class...which 99% of the class is unable to decipher. And then one kid raises his hand, and provides the appropriate answer. "BINGO!"...the teacher announces emphatically, as the bright student flashes a proud smile.

And then the topic is dropped and the class moves on to something else...because the rest of the class isn't "smart enough" to warrant pursuing that topic any further.

That's pretty much what went on in that thread...IMO.

[MJC922]

This one again eh. :)

As I recall the first problem was the thread title, as it turned out the results of the study had nothing to do with predicting a speed figure, it had to do with predicting the winner.

SJK and I went down the wrong (thread title) path of testing how well it predicted today's speed figure. As expected in that case the third race back is closer to today's figure than the 4th race back. The fact that 3rd back is slightly less predictive than the 4th race back I suppose means something profound to someone? I can verify that it is true.

In the end it comes down to finding what's most predictive not what isn't. And yes you can do better than the last race, in my case about 5% better.

[aaron]

Jim Cramer of HDW has projected speed ratings for every track. I don't know the exact % of winners,but I think it was about 32-33% and the roi was better than the takeout rate.

[MJC922]

Jim Cramer of HDW has projected speed ratings for every track. I don't know the exact % of winners,but I think it was about 32-33% and the roi was better than the takeout rate.

Might want to find out how many ifs ands or buts are involved to hit that number.

[traveler]

Might want to find out how many ifs ands or buts are involved to hit that number.

Cramers Projected SR - 10,000 race sample - ties bump number in the rank of 1, but 34.72% at $1.73 retrun per $2 bet. No ifs, lots of buts here...

WIN BETS
Field1 Field2 Starts Pays Pct $Net IV PIV HV
---------------------------------------------------------------
1 11,127 3,472 31.2 $1.73 2.50 1.06 1.33
2 9,050 1,811 20.0 $1.62 1.60 1.00 1.14
3 9,876 1,463 14.8 $1.60 1.19 0.97 1.04
4 9,976 1,118 11.2 $1.57 0.90 0.95 0.95
5 9,986 806 8.1 $1.50 0.65 0.89 0.84
6 9,992 603 6.0 $1.43 0.48 0.89 0.77
7 8,448 367 4.3 $1.34 0.37 0.85 0.69
8 6,258 204 3.3 $1.24 0.30 0.83 0.64
9 8,771 172 2.0 $1.08 0.21 0.74 0.53

Total 83,484 10,016 12.0 $1.48 1.00 0.95

[raybo]

Cramers Projected SR - 10,000 race sample - ties bump number in the rank of 1, but 34.72% at $1.73 retrun per $2 bet. No ifs, lots of buts here...

WIN BETS
Field1 Field2 Starts Pays Pct $Net IV PIV HV
---------------------------------------------------------------
1 11,127 3,472 31.2 $1.73 2.50 1.06 1.33
2 9,050 1,811 20.0 $1.62 1.60 1.00 1.14
3 9,876 1,463 14.8 $1.60 1.19 0.97 1.04
4 9,976 1,118 11.2 $1.57 0.90 0.95 0.95
5 9,986 806 8.1 $1.50 0.65 0.89 0.84
6 9,992 603 6.0 $1.43 0.48 0.89 0.77
7 8,448 367 4.3 $1.34 0.37 0.85 0.69
8 6,258 204 3.3 $1.24 0.30 0.83 0.64
9 8,771 172 2.0 $1.08 0.21 0.74 0.53

Total 83,484 10,016 12.0 $1.48 1.00 0.95

Hmmmm - looks like 31.2%, not 34.72%. Maybe I'm wrong though?

[Dave Schwartz]

Traveler,

Well, I certainly recognize that output.

Dave

[MJC922]

Cramers Projected SR - 10,000 race sample - ties bump number in the rank of 1, but 34.72% at $1.73 retrun per $2 bet. No ifs, lots of buts here...

WIN BETS
Field1 Field2 Starts Pays Pct $Net IV PIV HV
---------------------------------------------------------------
1 11,127 3,472 31.2 $1.73 2.50 1.06 1.33
2 9,050 1,811 20.0 $1.62 1.60 1.00 1.14
3 9,876 1,463 14.8 $1.60 1.19 0.97 1.04
4 9,976 1,118 11.2 $1.57 0.90 0.95 0.95
5 9,986 806 8.1 $1.50 0.65 0.89 0.84
6 9,992 603 6.0 $1.43 0.48 0.89 0.77
7 8,448 367 4.3 $1.34 0.37 0.85 0.69
8 6,258 204 3.3 $1.24 0.30 0.83 0.64
9 8,771 172 2.0 $1.08 0.21 0.74 0.53

Total 83,484 10,016 12.0 $1.48 1.00 0.95


Trust me I've been working with speed figures for over thirty years. If god comes to you and says he has a top fig last race that wins 34% you should call BS. In terms of accuracy the high-water mark for a number might be 27 without pace, with pace maybe a point or two higher. Top fig last race isn't exceeding 30% for anyone. People scam with this in two ways, first they like to use selective application, that is, they want to dig out a turf number for a turf race and ignore the last race if its dirt and / or more commonly they say the top ranked horse with an edge of X number of points wins 30+%, ignoring races when the top rank has a point edge which is small.

[traveler]

Trust me, I don't trust you at all based on your response. The figure presented is a Rank of number 1 for the PSR. Ray, your math is off too.

[MJC922]

Trust me, I don't trust you at all based on your response. The figure presented is a Rank of number 1 for the PSR. Ray, your math is off too.

What does ties bump number in the rank of one mean? How about we tie everyone, then we have 100% wins. Yipee!!

[Tom]

I asked a simple question..what is the purpose here?
What is the deliverable after all this math is done?
Do you end up with a list of horses in a race and the projected speed figure for each?

Is that really that unreasonable - take one race and post it?
I can only assume that the total lack of reply is that no one is actually doing it.

[classhandicapper]

If I knew what "W3 could be the minimum performance consistency point" means maybe I could answer some of my own questions, without input from TM or Magister Ludi. Maybe a Google of that might explain that phrase more?

Somewhere in the original thread someone explained to me that it is possible that w3 could be so correlated to w1 and w2 that you gain very little upside from including it. On the flip side going all the way back to w4 may start adding new information again that will be marginally helpful.

When I initially saw that w3 conversation I rejected it, but there is some possible logic in that response.

[classhandicapper]

I asked a simple question..what is the purpose here?
What is the deliverable after all this math is done?
Do you end up with a list of horses in a race and the projected speed figure for each?

Is that really that unreasonable - take one race and post it?
I can only assume that the total lack of reply is that no one is actually doing it.

Tom,

Many people focus on a horse's last race or most recent relevant race and base most of their thinking on that single race. Then they'll go from there as to whether the horse is improving, declining, suited to today's distance etc..

I fully recognize that most sets of PPs are a mixture of difference distances, surfaces, paces, trips, etc.. among horses at different points in their form cycle. So no one formula will capture it all. But I am still trying to outperform that "single race" approach. Surprisingly to me, it's not as easy as I thought.

My purpose is to have a more general understanding of how to weight races in the horse's PPs if you are using several relevant races in order to select more winners at potentially higher prices.

[aaron]

I asked a simple question..what is the purpose here?
What is the deliverable after all this math is done?
Do you end up with a list of horses in a race and the projected speed figure for each?

Is that really that unreasonable - take one race and post it?
I can only assume that the total lack of reply is that no one is actually doing it.
Tom,
You can go to HDW website or better yet call them. You will see all the horses in the race with a projected speed figure for each. As a handicapper,you will with agree with projection or disagree. it is really as simple as that.

[Tom]

I use HTR, so I am familiar with the projected figs.
I just was trying to understand what was being proposed here - was it like HTR does, or something different. Seems math guys are not good teachers.:rolleyes:

Btw, I do not use the projected figs, not even look at them.

I will assume there is nothing more to discuss here.

[thaskalos]

I will assume there is nothing more to discuss here.
Pretty good assumption...IMO.

[ReplayRandall]

Pretty good assumption...IMO.

From the start, this thread was DOA.....You can't resurrect nothing and think a miracle would occur, but some do.

[ebcorde]

there are just too many variables. Horses win by nose, neck, 1/2 length. a 72 speed figure beats an 82 by a neck.

and you find out the 18% Trainer of the 72 speed figure has his horse on the muscle bulking up, not the same Horse he was 2 months ago.

you need more data.

[classhandicapper]

Oh well, I guess I should be glad all you guys are surrendering. If I eventually find something other than looking at overall record does not add as much as some might imagine, I'll be one of the few that knows it. ;)

[ebcorde]

Oh well, I guess I should be glad all you guys are surrendering. If I eventually find something other than looking at overall record does not add as much as some might imagine, I'll be one of the few that knows it. ;)


not surrendering at all. it's just that no matter what you do with "numbers" you'll end up with the top picks

when I started in this game, I spent 2 weeks learning the form, the old equibase form. Then I tested myself for a few months. The old track equibase form always had their top 3 selections at the bottom of the page, so I would cover it up. I almost always ended up with the same picks they had.

Later on in life I handicapped solely using other methods, I came up with the same picks as when using speed.

Today most races I can handicap strictly on class alone If i want. However I use 4 different methods, and when all 4 methods produce the same horse, that type horse wins a lot of races.

I prefer the horse who meets or comes closest to meeting the metrics of all 4 methods I employ. so I don't get all worked up over speed/pace

I do use speed and Pace heavily. but also recent form , class and negative/positive situations for today's Trainers and Jockeys.

I posted for Tampa what Horses I would throwout based on my software.

[raybo]

Somewhere in the original thread someone explained to me that it is possible that w3 could be so correlated to w1 and w2 that you gain very little upside from including it. On the flip side going all the way back to w4 may start adding new information again that will be marginally helpful.

When I initially saw that w3 conversation I rejected it, but there is some possible logic in that response.

Nah, surely not, everyone knows that TM is full of crap and hot air. ;)

[aaron]

I have always used projections in my calculations and have adjusted my figures up or down based on various things I take into consideration.

[raybo]

Trust me, I don't trust you at all based on your response. The figure presented is a Rank of number 1 for the PSR. Ray, your math is off too.

I didn't do any math, just tried to read your post. Looks to me that the only column that looks like a hit percentage is 4th one, and that one is 31.2, not 34.72. I see 3,472 in the 3rd column but that definitely doesn't look like a percentage to me.

[Tom]

Try this....

[raybo]

Oh well, I guess I should be glad all you guys are surrendering. If I eventually find something other than looking at overall record does not add as much as some might imagine, I'll be one of the few that knows it. ;)

I'm with you! Please, the people who think this thread was "DOA", find another thread to rag. Yes, Tom, even you have the right to NOT post in a thread. You guys can totally ignore this thread.

We may not get anything of value in this thread, but there are hundreds of examples of that on this forum, and they don't get locked and they don't all get ragged on.

[raybo]

Try this....

Thanks Tom, but I still don't see a 34.72% hit rate anywhere, the highest is rank 1 at 31.2%.

[thaskalos]

Somewhere in the original thread someone explained to me that it is possible that w3 could be so correlated to w1 and w2 that you gain very little upside from including it.
Yes...it may well be that w3 could be discarded without having the final results suffer much adverse effect. But it could also be that, the minute you decide to exclude w3...that's when you'll miss the big payoff which would make your month.

When I first started in this game...I envisioned that I would discover a way of playing which would allow me to methodically grind consistent profits over time...without going through long "dry spells". What I found instead was that this game is very volatile...and the profits in it come in unpredictable waves. Yes...you can try to simplify things when you handicap, by eliminating this or that...trying to make the process more "logical". But this "simplification" often comes with a costly price-tag.

As more and more people mine these races with the help of vast databases...certain "rules" are bound to be created, which appear to simplify the handicapping process. But the horseplayer needs to be careful, IMO...because the profits in this game are often found in the EXCEPTION...and not the "rule".

[raybo]

not surrendering at all. it's just that no matter what you do with "numbers" you'll end up with the top picks

when I started in this game, I spent 2 weeks learning the form, the old equibase form. Then I tested myself for a few months. The old track equibase form always had their top 3 selections at the bottom of the page, so I would cover it up. I almost always ended up with the same picks they had.

Later on in life I handicapped solely using other methods, I came up with the same picks as when using speed.

Today most races I can handicap strictly on class alone If i want. However I use 4 different methods, and when all 4 methods produce the same horse, that type horse wins a lot of races.

I prefer the horse who meets or comes closest to meeting the metrics of all 4 methods I employ. so I don't get all worked up over speed/pace

I do use speed and Pace heavily. but also recent form , class and negative/positive situations for today's Trainers and Jockeys.

I posted for Tampa what Horses I would throwout based on my software.

That highlighted statement is patently false. I use nothing but numbers and logic decision trees in my program and I get selections, and winners, from all odds ranges, from 1/9 to triple digits.

[raybo]

Yes...it may well be that w3 could be discarded without having the final results suffer much adverse effect. But it could also be that, the minute you decide to exclude w3...that's when you'll miss the big payoff which would make your month.

When I first started in this game...I envisioned that I would discover a way of playing which would allow me to methodically grind consistent profits over time...without going through long "dry spells". What I found instead was that this game is very volatile...and the profits in it come in unpredictable waves. Yes...you can try to simplify things when you handicap, by eliminating this or that...trying to make the process more "logical". But this "simplification" often comes with a costly price-tag.

As more and more people mine these races with the help of vast databases...certain "rules" are bound to be created, which appear to simplify the handicapping process. But the horseplayer needs to be careful, IMO...because the profits in this game are often found in the EXCEPTION...and not the "rule".

Not being argumentative, on purpose, but wouldn't you agree that universally ignoring the 3rd race back is an EXCEPTION.. and not the "rule"?

[thaskalos]

Not being argumentative, on purpose, but wouldn't you agree that universally ignoring the 3rd race back is an EXCEPTION.. and not the "rule"?
No. If your database tells you that you could afford to ignore w3...then that becomes your newly-discovered "rule". I don't believe in ignoring ANYTHING. What possible determination could I make about w3...without carefully considering the circumstances of that race? Why would I reject w3...while still keeping w4?

The order in which these races appear on the page should be the main determinant when deciding which race we keep...and which race we throw away?

[raybo]

No. If your database tells you that you could afford to ignore w3...then that becomes your newly-discovered "rule". I don't believe in ignoring ANYTHING. What possible determination could I make about w3...without carefully considering the circumstances of that race? Why would I reject w3...while still keeping w4?

The order in which these races appear on the page should be the main determinant when deciding which race we keep...and which race we throw away?

Whew... I know what you're saying Gus, and I don't disagree, but universally ignoring a particular race based on recency ranked order would definitely be an exception to the rule in racing. Just because one person decides to use that method, doesn't now make it a rule in racing, for everyone else. I seriously suspect that most people who think that stepping out of the box of traditional experience is becoming more and more important in order to make profit in racing, would agree that universally ignoring the 3rd race back is indeed stepping out of that box, and is an exception.

Do I do that? No, I don't, but then I haven't done that research either. Mostly because I'm still doing fine and I don't use pace or speed figures anyway.

[thaskalos]

Whew... I know what you're saying Gus, and I don't disagree, but universally ignoring a particular race based on recency ranked order would definitely be an exception to the rule in racing. Just because one person decides to use that method, doesn't now make it a rule in racing, for everyone else. I seriously suspect that most people who think that stepping out of the box of traditional experience is becoming more and more important in order to make profit in racing, would agree that universally ignoring the 3rd race back is indeed stepping out of that box, and is an exception.

Do I do that? No, I don't, but then I haven't done that research either. Mostly because I'm still doing fine and I don't I use pace or speed figures anyway.

In his original thread here...TrifectaMike suggested that there was a way of accurately predicting the horse's upcoming speed figure, by assigning different weights to the horse's prior 4 speed figures. Greyfox asked him in a subsequent post if these prior 4 races shouldn't be examined for class and distance concerns...before these weights are assigned. TrifectaMike casually brushed that suggestion off...as if it were an unimportant consideration. "Let's move on"...TM said.

Move on to WHAT?

We are going to ignore the class, surface and distance of the horse's prior 4 races...and we are going to endeavor to predict the horse's upcoming speed figure...by assigning different weights to those 4 prior races. And then what? We march off to the betting window...and confidently put our money down on our choice?

We are not razzing the thread, Ray. We are trying to UNDERSTAND it. But someone has to explain the basic premise here...and it's obvious that TM won't be the one to do it.

Will YOU?

[thaskalos]

I seriously suspect that most people who think that stepping out of the box of traditional experience is becoming more and more important in order to make profit in racing, would agree that universally ignoring the 3rd race back is indeed stepping out of that box, and is an exception.


Again...I disagree. If our site here is indicative of the public opinion at large, I submit that the vast majority of the players have already IGNORED the horse's 3rd race back...so, CONSIDERING this race would seem to be the "exception". The majority of the posters here have expressed the opinion that they are even unwilling to go beyond the horse's most RECENT race...unless there was a "valid reason" to do so.

[raybo]

In his original thread here...TrifectaMike suggested that there was a way of accurately predicting the horse's upcoming speed figure, by assigning different weights to the horse's prior 4 speed figures. Greyfox asked him in a subsequent post if these prior 4 races shouldn't be examined for class and distance concerns...before these weights are assigned. TrifectaMike casually brushed that suggestion off...as if it were an unimportant consideration. "Let's move on"...TM said.

Move on to WHAT?

We are going to ignore the class, surface and distance of the horse's prior 4 races...and we are going to endeavor to predict the horse's upcoming speed figure...by assigning different weights to those 4 prior races. And then what? We march off to the betting window...and confidently put our money down on our choice?

We are not razzing the thread, Ray. We are trying to UNDERSTAND it. But someone has to explain the basic premise here...and it's obvious that TM won't be the one to do it.

Will YOU?

To my knowledge of the thread, TM never said that there was an accurate way of projecting speed figures, he even went on late r to say that he uses a hierarchical Bayes Model to project speed figures. The "story" he posted was just an exercise, IMO, something to get a conversation going, and in TM's typical MO act as the professor teaching a class. That's fine and good with me, there are other posters who are worth conversing with. Al;so, he never said that the speed figure projection example would be of any value at all, and had nothing to do with value.

Some people just overstayed themselves because they really expected TM to offer an answer, which he almost never does.

"Let's move on" was just his way of getting to the next part of his "story". He's not the only one here who illicit participation in threads that lead nowhere. His moniker alone should be enough to warn some here from participating in the thread, but they do it anyway, maybe they get something out of it, I don't know.

[classhandicapper]

But someone has to explain the basic premise here...and it's obvious that TM won't be the one to do it.

Will YOU?

I think the premise may be that a lot of the things we "think" are so when it comes to handicapping come from intuition and experience. But perhaps some aren't accurate. Our minds tend to correlate and confirm things we think going in.

So in the case of w3 or in some of the cases I have tested, the conventional wisdom may not be true.

I'm not sure what to with that w3 information.

But if I was evaluating an older horse with relative stable form and I knew that weighing the last 3 races at 58%, 32% and 10% was better than just looking at the last race or using some other weight, I might make better value oriented decisions. On the flip side, if I found that looking at all 3 races was making my decision making more complex but not adding anything over the long haul relative to looking at just the last race, I could adjust my thinking in the opposite way.

I argued in the original thread that there are no formulas that will work all the time. I still think that. But there may be some general principles that will help.

[raybo]

Again...I disagree. If our site here is indicative of the public opinion at large, I submit that the vast majority of the players have already IGNORED the horse's 3rd race back...so, CONSIDERING this race would seem to be the "exception". The majority of the posters here have expressed the opinion that they are even unwilling to go beyond the horse's most RECENT race...unless there was a "valid reason" to do so.

I don't know about the majority here, but I certainly don't, and most of the posters that I find most knowledgeable about the game don't either. Most people here that I have respect for, as handicappers, don't use speed figures in isolation either, it's just one of the tools and analysis they use. So, what TM was doing would not have been anything of tremendous importance anyway, even if one could actually project speed figures from the last 4 figures. Maybe it can be done, don't know, I use all surface and distance qualified pacelines for my gradings, regardless of class, time off, distance changes, trainer moves, jockey stats, or anything else. Works fine for me. So, who am I to say that projecting a speed figure from the last 4 figures won't work, or can't be done?

[thaskalos]

Well...I am in agreement with Tom in this case. As is usually the case when viewing a TM thread...I think my time could be better spent. I leave this topic to YOU guys...and I wish you luck.

[Cratos]

I think the premise may be that a lot of the things we "think" are so when it comes to handicapping come from intuition and experience. But perhaps some aren't accurate. Our minds tend to correlate and confirm things we think going in.

So in the case of w3 or in some of the cases I have tested, the conventional wisdom may not be true.

I'm not sure what to with that w3 information.

But if I was evaluating an older horse with relative stable form and I knew that weighing the last 3 races at 58%, 32% and 10% was better than just looking at the last race or using some other weight, I might make better value oriented decisions. On the flip side, if I found that looking at all 3 races was making my decision making more complex but not adding anything over the long haul relative to looking at just the last race, I could adjust my thinking in the opposite way.

I argued in the original thread that there are no formulas that will work all the time. I still think that. But there may be some general principles that will help.
Raybo, I have re-read TrifectaMike’s original post and I have now come to the conclusion that he being a Bayesian would’ve probably suggested to start with a high probability region, or credibility interval which represents the potential values based on previous evidence.

The high probability region represents the range of reasonable values for the weights. It is not the frequency of anything; not even approximately. It describes an uncertainty range for the unknown, but fixed numbers in the data and its purpose is to generate error values used in the sensitivity analysis.

This is a very mathematical exercise.

However it is amazing to me that when certain posters don’t understand the thesis of a post they go off with unwarranted sarcasm and cynicism.

Therefore I would suggest for those who think that TrifectaMike’s posting behavior is Socratic they should read (if they haven’t) E.T. Jayne’s “Probability Theory: The Logic of Science.”

[raybo]

I think the premise may be that a lot of the things we "think" are so when it comes to handicapping come from intuition and experience. But perhaps some aren't accurate. Our minds tend to correlate and confirm things we think going in.

So in the case of w3 or in some of the cases I have tested, the conventional wisdom may not be true.

I'm not sure what to with that w3 information.

But if I was evaluating an older horse with relative stable form and I knew that weighing the last 3 races at 58%, 32% and 10% was better than just looking at the last race or using some other weight, I might make better value oriented decisions. On the flip side, if I found that looking at all 3 races was making my decision making more complex but not adding anything over the long haul relative to looking at just the last race, I could adjust my thinking in the opposite way.

I argued in the original thread that there are no formulas that will work all the time. I still think that. But there may be some general principles that will help.

I think, generally, by the time the regression process gets down to the 3rd and 4th races, the weightings are so close to each other, and the weightings compared to the last and 2nd race back are so low, that the 3rd race back may well not add anything significant to the process. The same may apply to the 4th race back too, if you use the 3rd race back instead.

I suspect the weightings would be something like 6 2 1 1 rounded off, or similar. From that statement of TM's about the 3rd race back being the least significant, I take it that the majority of the time one of the last 2 races are the most significant, with the 4th having some small degree of significance long term (probably due to normal variance within that set of data), if you are using a paceline selection methodology. Sure, there will be times that the 3rd race back is more applicable, in the eyes of the handicapper, but there is also the possibility that using that figure will not help (or hurt) your hit rate, long term.

Of course, the purpose of the regression was to enable the projection of a future speed figure that would point to winners, not necessarily profits. The profits portion, as we know, is a separate function and we often don't bet the horse who has the highest probability of winning (best projected speed figure in this case), or the highest ranking after our other analysis anyway. That exercise was based, not on profit potential, but on projecting a future speed figure, and ultimately, a better rating for identifying probable winners, long term. Perhaps the end goal was to create an odds line or probability line, I don't know. But, if in fact that regression analysis really said that the 3rd race back was insignificant, then for that sample of races, and for that full population of horses, and for that full population of tracks, it would be ill advised to dispute it, IMO, unless one has done the same analysis, of the same data set, and come out with a different observation. I will accept it as true, for that sample, regardless of my own personal beliefs.

[Gamblor]

If the answer is that there is some sort of cyclicity and the third race back is the "minimum performance in the pattern" or whatever the post said, then my question is this: for the horse's next start, does that previous third run back (which is now their fourth run back) suddenly assume more importance than the run after it, when initially it was irrelevant in comparison?

[storyline]

After reading this thread from the beginning I just shake my head. I almost wouldn't bet a race without making projected figs for every runner.

TM whom I don't know is most likely playing with you, I certainly doubt it's as silly as looking at the best of the last four races or anything like that.

Serious question: how do you separate your contenders?

[Gamblor]

After reading this thread from the beginning I just shake my head. I almost wouldn't bet a race without making projected figs for every runner.

TM whom I don't know is most likely playing with you, I certainly doubt it's as silly as looking at the best of the last four races or anything like that.

Serious question: how do you separate your contenders?

Thing is, as a few have mentioned earlier ITT and the prior one, the question by TM wasn't actually anything remotely about "which horse will win." It was one of those purely mathematical questions designed to analyse the method and improve it, a question which had no regard whatsoever for whether or not the answer would prove profitable.

[thaskalos]

If the answer is that there is some sort of cyclicity and the third race back is the "minimum performance in the pattern" or whatever the post said, then my question is this: for the horse's next start, does that previous third run back (which is now their fourth run back) suddenly assume more importance than the run after it, when initially it was irrelevant in comparison?

I asked this question earlier in this thread. Why reject the horse's 3rd-race back...when you are keeping the horse's 4th race back? Why is the 4th-race back more important than the race preceded It?

And...how can you accept the speed figure of a race...without noting the class, distance and footing of that race...to see If these conditions apply to the race TODAY?

[Tom]

If today's race is a turf route, the third back an off the turf race, and the fourth back a turf route, I would not give the third back any value at all.

[Gamblor]

I asked this question earlier in this thread. Why reject the horse's 3rd-race back...when you are keeping the horse's 4th race back? Why is the 4th-race back more important than the race preceded It?

And...how can you accept the speed figure of a race...without noting the class, distance and footing of that race...to see If these conditions apply to the race TODAY?

For the purposes of this exercise I can understand ignoring those things. It would be foolish to do it in reality, but from a theoretical perspective it's fine.

Suggesting run 3 is less important than run 4 though, then suddenly saying it assumes more importance after they've had one MORE run, however, makes no sense, and I've seen nothing to argue the case for it.

[Gamblor]

If today's race is a turf route, the third back an off the turf race, and the fourth back a turf route, I would not give the third back any value at all.

Correct, and I've read you argue the "what is the point?" argument ITT from start to finish, and while you're right I'll respectfully say that your argument doesn't matter. This is a theoretical discussion only, not a practical one. You're looking for practical answers to a theoretical question.

[thaskalos]

"Theory" is good...but "practice" is better.

[CincyHorseplayer]

"Theory" is good...but "practice" is better.

Right why theorize when you can play. If there is some remarkably negative handicapping transformation that occurs between these 2 realities=you got a problem!

[Gamblor]

Theorising is a perfectly fine thing to do, because it's the best way to make new discoveries. Eventually yes you have to put your theories to some sort of practical use, but starting with theory is OK.

[gm10]

Trust me I've been working with speed figures for over thirty years. If god comes to you and says he has a top fig last race that wins 34% you should call BS. In terms of accuracy the high-water mark for a number might be 27 without pace, with pace maybe a point or two higher. Top fig last race isn't exceeding 30% for anyone. People scam with this in two ways, first they like to use selective application, that is, they want to dig out a turf number for a turf race and ignore the last race if its dirt and / or more commonly they say the top ranked horse with an edge of X number of points wins 30+%, ignoring races when the top rank has a point edge which is small.

I agree with the "top last speed" win % plateauing around 29-30%, but projected speed ratings that are based on previous SR's which are weighted according to days ago/surface/going, do go to up 35% in my experience.

[gm10]

The reason I resurrected this thread is because I've done some more studies since then. We all know that occasionally a horse's last race has to be dismissed or discounted, but the thing I am finding is that it's fairly difficult to outperform a horse's last race alone by trying to combine multiple races in a systematic way.

I even tried some approaches I intuitively thought should work, but didn't.

For example, I looked at a horse's last race, best of last 3 and best of last 5.

I assumed that if I combined best of last 3 or best of last 5 (which performed reasonably well on their own) with last race and weighted it so that they were only strong enough to serve as a tiebreaker when 2 horses were otherwise very similar off their last race, it should outperform last race alone. But it did not. I tried quite a few things with various weights and nothing systematic worked. The results would be similar, but last race alone would eek out a slightly better result.

I'm starting to think you need a very good reason to look much past a horse's last race. Combining races may find winners that last race won't get, but they will just be different horses. There won't necessarily be more of them or better values.

Try this:

- Give weights to each PP in function of recency
- Give weights to each PP in function of today's surface
- Give weights to each PP in function of today's going
- Give weights to each PP in function of today's distance

This will give you four "weighted averages". Average these four numbers. As long as your weighting was done sensibly, this average of averages will outperform the "LTO speed figure" angle.



For "recency", I just use this formula:

weight = exp(-0.01 * days_since_last_run)

If a race in the PP's was 10 days ago, the weight for it is 90%. If another race was 70 days ago, the weight drops to 50%. (Plot on an excel sheet to see all values)

Suppose these are the only 2 PP's. The "weighted recency rating" is thus calculated:

(0.90 * SR1 + 0.50 * SR2) / (0.90 + 0.50)

[classhandicapper]

I asked this question earlier in this thread. Why reject the horse's 3rd-race back...when you are keeping the horse's 4th race back? Why is the 4th-race back more important than the race preceded It?

And...how can you accept the speed figure of a race...without noting the class, distance and footing of that race...to see If these conditions apply to the race TODAY?


I think most would agree that you should look at all the PPs (other than perhaps system players and some computer model players).

The question I was getting at was that if you could control for all the complexities (or if all of a horses last 4-5 races were basically earned under similar conditions), you'd still want to know how important a horse's last race was relative to the previous and so on back. That kind of knowledge would also be useful even if the PPs were a little messier.

I continued studying this a bit and more or less stand by my original assertion.

There are situations where a horse's last race can and should be dismissed, but I don't see a lot of value in going back into the PPs past the last race and weighing various races as long as that last race occurred under today's conditions with no extreme pace/trip/bias/track condition etc... issues to consider. When I've tried to use multiple races it varied the horses I was selecting, but it didn't vary the results much.

The bigger skill seems to be in evaluating when a bad last race (or 2) was entirely the result of the conditions/trip and when it was just a bad race.

For example, we have a horse that typically runs in the mid 90s. He runs on a sloppy track, runs very poorly, and puts up a 75.

Was it the slop or did he go off form and it just happen to be a sloppy track that day? Maybe it was a little of both?

Substitute turf for slop.

Substitute raced on a bad rail.

Answering questions like that accurately will help you outperform last race.

[aaron]

What do you do if a horses race 2 back was 20 points higher than his normal race,but there is nothing to indicate the number is incorrect or aided by a bias or a setup. Would you include that number in your calculations. I have generally thrown out that number unless it was a young and lightly raced horse. In most situations these horses are one number horses,not to be repeated

[thaskalos]

I think most would agree that you should look at all the PPs (other than perhaps system players and some computer model players).

The question I was getting at was that if you could control for all the complexities (or if all of a horses last 4-5 races were basically earned under similar conditions), you'd still want to know how important a horse's last race was relative to the previous and so on back. That kind of knowledge would also be useful even if the PPs were a little messier.

I continued studying this a bit and more or less stand by my original assertion.

There are situations where a horse's last race can and should be dismissed, but I don't see a lot of value in going back into the PPs past the last race and weighing various races as long as that last race occurred under today's conditions with no extreme pace/trip/bias/track condition etc... issues to consider. When I've tried to use multiple races it varied the horses I was selecting, but it didn't vary the results much.

The bigger skill seems to be in evaluating when a bad last race (or 2) was entirely the result of the conditions/trip and when it was just a bad race.

For example, we have a horse that typically runs in the mid 90s. He runs on a sloppy track, runs very poorly, and puts up a 75.

Was it the slop or did he go off form and it just happen to be a sloppy track that day? Maybe it was a little of both?

Substitute turf for slop.

Substitute raced on a bad rail.

Answering questions like that accurately will help you outperform last race.

A more realistic (and more perplexing) example is the horse that has run a couple of sharp races in a row, earning contending figures...but has followed that up with a lackluster last race, where no reasonable excuse for the sub-par effort could be detected. Does the handicapper focus on the sub-par last race...to the exclusion of the two sharp efforts that preceded it? If the unimpressive last race took place at the same class, surface and distance as today's race, and no race "dynamic", or surface "condition", could be blamed for the horse's sub-par effort...do we just conclude that the horse is "tailing off form"...and expect another sub-par effort TODAY?

When we say that we "handicap"...are we looking at the totality of a horse's work...or are we just fixating on its most recent "representative race"?

[classhandicapper]

A more realistic (and more perplexing) example is the horse that has run a couple of sharp races in a row, earning contending figures...but has followed that up with a lackluster last race, where no reasonable excuse for the sub-par effort could be detected. Does the handicapper focus on the sub-par last race...to the exclusion of the two sharp efforts that preceded it? If the unimpressive last race took place at the same class, surface and distance as today's race, and no race "dynamic", or surface "condition", could be blamed for the horse's sub-par effort...do we just conclude that the horse is "tailing off form"...and expect another sub-par effort TODAY?

When we say that we "handicap"...are we looking at the totality of a horse's work...or are we just fixating on its most recent "representative race"?

Your last question is exactly the question I am asking and your example a perfect one.

For decades I would focus on the body of work and sort of mentally weigh all the information based on experience, giving more weight to the most recent races. The idea for the study was to improve on "mentally weigh" and try to get some concrete answers to these questions.

A horse with few sharp races before a bad one is certainly more likely to jump back up than a horse without any good races like that, but he's not as likely to run back to them as a horse whose last race was good.

[classhandicapper]

Try this:

- Give weights to each PP in function of recency
- Give weights to each PP in function of today's surface
- Give weights to each PP in function of today's going
- Give weights to each PP in function of today's distance

This will give you four "weighted averages". Average these four numbers. As long as your weighting was done sensibly, this average of averages will outperform the "LTO speed figure" angle.



For "recency", I just use this formula:

weight = exp(-0.01 * days_since_last_run)

If a race in the PP's was 10 days ago, the weight for it is 90%. If another race was 70 days ago, the weight drops to 50%. (Plot on an excel sheet to see all values)

Suppose these are the only 2 PP's. The "weighted recency rating" is thus calculated:

(0.90 * SR1 + 0.50 * SR2) / (0.90 + 0.50)


That seems like a good approach.

[thaskalos]

Your last question is exactly the question I am asking and your example a perfect one.

For decades I would focus on the body of work and sort of mentally weigh all the information based on experience, giving more weight to the most recent races. The idea for the study was to improve on "mentally weigh" and try to get some concrete answers to these questions.

A horse with few sharp races before a bad one is certainly more likely to jump back up than a horse without any good races like that, but he's not as likely to run back to them as a horse whose last race was good.

I don't particularly care if the horse with the sharp last race runs back to it more often than the horse who ran uncharacteristically bad last out...because, as a bettor, my battle is against the odds. And the horse who ran uncharacteristically bad last out often offers enough betting value to overcome his apparent "inconsistency".

The horse with the sharp last race has the obvious disadvantage of having that race out there for everybody to see.

[classhandicapper]

I don't particularly care if the horse with the sharp last race runs back to it more often than the horse who ran uncharacteristically bad last out...because, as a bettor, my battle is against the odds. And the horse who ran uncharacteristically bad last out often offers enough betting value to overcome his apparent "inconsistency".

The horse with the sharp last race has the obvious disadvantage of having that race out there for everybody to see.

No doubt you will get higher average prices by throwing out a horse's last dull looking race and looking further back. But the public also knows that horses with good back races are more likely to bounce back with another good effort today. So people bet accordingly.

In order to evaluate the values correctly, you have to know the actual probability of that happening.

IMO there are a lot of factors that matter. But trying to quantify it so you make good value judgments is a lot tougher than knowing the generalities that the public also knows.