Choosing Pacelines

[sq764]

I was curious how everyone decides on paceline selection.. I think this is the most difficult part of my game and possibly the same for others..

I try to choose the most recent paceline, as much as possible, barring any 'out of the norm' performances for the horse..I typically will go back 3-4 races at most..

My biggest issues come when a horse shows a big speed fig last out, after 4-5 average figs.. I am always inclined to ditch the big fig and to use a more typical line for a horse.. Mind you, these are mostly older claimers with a lot of races under their belt, not an improving, lightly raced 2 or 3 yo..

Anyone have a more rigid criteria for paceline selection? I would be very curious to hear..

[cj]

I would never try to judge a horse on one paceline, I try to look at EVERY paceline. This is the downfall of most software in my opinion. There is no "usual" performance, only the one the horse is going to give today that matters. Even if you could somehow measure form, and say this horse is 100% dead fit and healthy, and the last time he was in this exact condition, he did such and such, it doesn't mean he will do it today. You still have to look at pace scenarios, running style, distance and maybe surface changes, etc, etc.

[jfdinneen]

sq764,


You raise an interesting point on the difficulty of paceline selection particularly when the most recent paceline is unusually fast or slow.

Whatever selection criteria you favor, your instinct to edge towards the median (average) performance rather than selecting one of the outliers (high or low) is the statistically correct approach.

To offset the impact of such outliers, I use the average of the last three similar pacelines - distance (+/- 0.5f), condition (firm, good) - equivalent to using a moving average with stock prices.


Best wishes,


John

[stuball]

JFD

WHEN YOU TAKE LAST 3 AVE --DO YOU CONSIDER CHANGES TRAINER,

BLINKS, MED ???

STUBALL

[andicap]

I've posted a few of my hair-brained theories before and here are the links

http://www.paceadvantage.com/forum/showthread.php?p=128693&highlight=paceline#post128693


here's a long thread on this from april 2004 (with links to two other extensive discussions)

http://www.paceadvantage.com/forum/showthread.php?t=11166&highlight=pacelines

This discussion of Master Magician also had a lot on PL selection recently

http://www.paceadvantage.com/forum/showthread.php?t=14227&highlight=pacelines

http://www.paceadvantage.com/forum/showthread.php?t=12797&page=1&highlight=pacelines

http://www.paceadvantage.com/forum/showthread.php?t=10966&highlight=pacelines

That should keep you busy for a while

:D

[jfdinneen]

Stuball,

I should have added to my original submission that I compare the short term moving average (last 3) to the longer term moving average (last 5/7) to monitor upswings and downswings in form - equivalent to how a Wall Street analyst monitors stock price movements!

Because I focus exclusively on median (average) performance, I do not need to compensate for all the many factors that might impact an individual paceline. It does mean, however, that in general I tend to concentrate on older horses with longer track records.

Best wishes,

John

[andicap]

JF,
Median or average? Which one do you use? There's a difference, as you know.

[jfdinneen]

Andicap,


Good point.

I always use the Median (middle performance) to minimize the negative impact of outliers (high or low) - I put the word average in brackets for those readers who are not sufficiently comfortable with statistical concepts.


Best wishes,


John

[Larry Hamilton]

a comment bout variants: If there is a race factor today about which you have a measurement from a sample size of say 500 which has an output of 45 somethings. Your computer program says the days races indicate a variant decline of say 3 somethings. So, you decide that your measurement of 45 for the day should be degraded to 42. One can only conclude from this conduct of adjusting the output of a large sample from a small one that your confidence in your original calculation is suspect--why bother? What you are doing is squirming to the noise of your calculation.

as to average last 3 or last running lines--How do you know the "running average" is any more predictive than any other method. (A question meant for discussion not antagonism)

[so.cal.fan]

I don't like to go back too far.......I tend to go along with sq764.
I will look at the last three or four races......skipping races on different surfaces or sprints/routes.
A race too far back nowadays, is usually not as reliable as it used to be for whatever reasons.

[sq764]

Here's my biggest issue... Let's say you have the following pacelines:

Date/finish/LB/Speed fig (Assuming all at 6F on fast track)

1/28/05 - 5th/5/72
1/14/05 - 6th/7/65
11/14/04 - 1st/0/84
10/27/04 - 3rd/2/80


Which one of these pacelines would you use for a race on 2/10/05??

[cj]

That scenario, all things being equal, I'd go with the 80. I see a 60 day layoff, a return that was mediocre, some improvement next out, and probably real sharp for this race. I'd like to see a few more past races to assess past ability, but given what you listed, I'd go with the 80.

[Overlay]

Data I've seen has suggested that an average (mean) of figures from those races out of the horse's last three which were run on the same surface (dirt or turf, irrespective of condition) as today's race, or from all of the last three races (if none were run on the same surface as today), has the greatest predictive power.

[sq764]

Data I've seen has suggested that an average (mean) of figures from those races out of the horse's last three which were run on the same surface (dirt or turf, irrespective of condition) as today's race, or from all of the last three races (if none were run on the same surface as today), has the greatest predictive power.
Hmm, so what about in this scenario:

1/30/05 - 6F - 5th/ 70 speed fig
1/12/05 - 6F - 1st/90 speed fig
12/28/04 - 6F - 4th/72 speed fig
12/14/04 - 6F - 6th/73 speed fig


You would average the 70,90 and 72?

See, in this scenario, Iwould toss the 90

[Overlay]

I would certainly understand your reasoning and logic in tossing out that 90 as an atypical outlier. But, on the other hand, what basis would there have been, working with the other figures that you listed, for predicting that the horse would jump up to a 90 in the second race back (which was the only race of the four which it won)? It seems to me that, if you don't take that 90 into consideration, you're discounting the horse's demonstrated potential for the kind of in-and-out, sudden improvement it showed in that race. (Just my opinion.)

[ratpack]

I have 2 points to make and please don't take this as empirical evidence just as information that I have gleaned.

1. I have found that using only using 1 Paceline whatever your selection method is is far better than using an average of 2 or 3 different pacelines.

2. I also agree with Mike Pizolla in his book to use the best available Paceline for TURF RACES.

[cj]

I don't understand the averaging. Maybe when the horse feels good, he runs 90, otherwise, 70, so what does an average of 80 mean. Fun with numbers won't win in this game, its all about the analysis of numbers, especially as it pertains to current form.

[ratpack]

I agree when I was using Equsim they had a avg function that when I tried it never had a better win% than a single paceline

[andicap]

I can see a three-race moving average as an indicator of the horse's form cycles maybe.

[Overlay]

cj:

I agree with you, and to me your point emphasizes my original question. Without factoring in the 90, what basis is there for projecting a repetition of the kind of sudden improvement the horse showed on January 12 (in sq764's example). And whether that would be considered mere playing with numbers or not, isn't this thread about using past quantitative values to approximate or predict future performance? My main concern in evaluating a performance measure is the predictive power it exhibits, and whether it does so over a sufficient range of circumstances and number of races to suggest that its performance is not the product of a statistical fluke, or of interaction with other handicapping variables. If any criterion (including the three-race average I mentioned) passes those tests, then that satisfies me as to its potential usefulness to my handicapping. As others have noted on this board, one of the great things about this game is the variety of approaches a person can use to arrive at valid conclusions and blend them into an overall profitable approach. Also, I concede that I could probably improve my bottom line if I were willing to devote more time to detailed race-by-race qualitative analysis of the significance of each past figure for the horse's likely performance today (akin to Steve Davidowitz's question, "What is he doing in today's race?", or to the type of race-by-race judgments in assigning speed figures that Andy Beyer discussed at length in Picking Winners). But since studies of that nature have already been conducted, I prefer to avail myself of that information and use measurement techniques which have stood up under wide-ranging examination (even if it does involve an incremental decrease in thoroughness of analysis) rather than spend the time that would be required for horse-by-horse decisions about the proper pace line to use, which to me has its own pitfalls in the subjective judgments that it requires.

[sq764]

I think this all leads to the question of whether we are trying to find

a) What the horse is capable of, the absolute best race he can run, or

b) What type of race the horse will most likely run today

[cj]

I think you need to find both. Not 1 or 2, but 1 AND 2.

[LARRY GEORGE]

WHEN I USE TO GO TO THE SARTIN CLASSES IN BEAUMONT CALIF.
HE SAID TO USE THE BEST RACE OF THE LAST THREE RACES AT A
SIMILAR DISTANCE AND SURFACE AND NOT USING A MAIDEN WIN
HE ALSO DID A FEW THINGS TO THE TRACK VARIANT AND SPEED
RATINGS.
GET IN TOUCH WITH ME AND MAYBE I CAN EXPLAIN IT TO YOU :confused:

[sq764]

That's pretty much what I do when I use MPH and its been very good to me..

Was just curious how people handle those unique situations.

[LARRY GEORGE]

I JUST RAN INTO TOM BROHAMER AT LOS ALAMITOS HE SAID HE'S NOT
PLAYING AS MUCH DUE TO SUCH LOW PRICES BUT HE'S STILL AT IT. :)

[sq764]

I JUST RAN INTO TOM BROHAMER AT LOS ALAMITOS HE SAID HE'S NOT
PLAYING AS MUCH DUE TO SUCH LOW PRICES BUT HE'S STILL AT IT. :)
PMTR still by far the best book I have ever read..

[46zilzal]

I JUST RAN INTO TOM BROHAMER AT LOS ALAMITOS HE SAID HE'S NOT PLAYING AS MUCH DUE TO SUCH LOW PRICES BUT HE'S STILL AT IT.
Love Los Alamitos (Old Kaweah Bar was a great one:The Palimino Freight Train) my all time second favorite course behind Anita, but

Stay away from sheets players, and the folks with laptops and find those little venues, with no name, cheap horses and the prices GO WAY UP. Housewifes must be the staple bettor at may of these because the odds are wildly irregular. You have to change your ideas, but the waters are less cluttered.

Cheapies have SHORT, definitive cyles, are stuck in a rut of a particular racing style (and are very HARD PRESSED to adapt to anything different: like some people I know), trainers are hungry as are the riders.

I have found several that are downright amazing.

[DJofSD]

One approach I use to select pace lines is looking for patterns.

Sometimes it is as simple as, if today is the 3rd start after a freshening or lay off, go back to the last time this happened and use the line that corresponds to todays race. Barring other factors, that's likely how the horse will run today.

Doc often preached pattern recognition. Horses are very much creatures of habit.

Broke its maiden in its 1st race? Has the horse fired fresh before? Why not go back to that line 6, 7 or 8 lines back if it is a good effort under similar circumstances?

Patterns, baby, it's patterns.

But it takes guts to go back, back, way back to use a pace line that's old and crusty. I'd been chastised more than once (even in writing) by Doc for doing that. Yes, Doc said many times, use the best of the last three. But that was for those still think inside the box.

DJofSD

P.S. Isn't Tom Brohamer living in P. S. any more? Los Al is a long way to go when you have multiple indian casinos much closer. You can bet if he's not playing the races as much now, it's because he's out on the golf coarse. And, if you see him again, ask him when is he going to eat that hat he promised to consume about 12 or 13 years ago while in L. V. during one of the seminars if his selection did not win<BG>.

[thoroughbred]

I have always tried to avoid using averages. After all the average almost always does not represent any of the races that went into the average calculation. Each of the races that went into the calculation of average is different from the average values calculated.

So what do we learn when we compare the average performance of the horses in today's race. Nothing very useful, I believe. Because each horse will run differently today than its average value.

I believe it is more important to look for things like:
What was the best the horse has ever done? How does this compare with the best of the other horses?
Is there evidence of a trend either up or down in recent performances that will let us estimate if the horse, or horses can repeat their best or, perhaps do even better? Or is the trend in the worse direction?

Those, are examples of more meaningful questions than what are the averages?

Here is a simple "thought" experiment to make the point more clear.
We have two horses who have each raced twice before at 6 furlongs.
One hores raced at 68 and 70 seconds. The other at 67 and 71 seconds. Both have the same 69 second average. What can we expect if they raced against each other today? A tie? Unlikely.

"Rolling" the averages doesn't help either because of the up and down variations of the horses from race to race.

[46zilzal]

Here is a simple "thought" experiment to make the point more clear.
We have two horses who have each raced twice before at 6 furlongs.
One hores raced at 68 and 70 seconds. The other at 67 and 71 seconds. Both have the same 69 second average. What can we expect if they raced against each other today? A tie? Unlikely.

However "tandems" occur all the time (not specifically from races as widely divergent as your example). Trainers sends the same horses to face each other again: What do you think the winning connections will do? More of the same. It worked before it should work again. The OTHERS: change the modus operandi, challenge the winner in a differnt way and it is OFTEN the difference.

Bet Twice over Alysheba to note a classic example, but see it at EVERY level. Races are MUCH more than time of course.

[Shacopate]

SQ,

I don't handicap like this anymore. But in the past, I chose the most recent paceline that best fit todays distance and class. And found the best value when the horse was recently in over it's head and the paceline came from the 3rd or 4th back without a lenghty layoff.

Your first example; IMHO---shows a horse with a postive form cycle and ready for a peak effort. New top, bounce, recovery close to middle. This was one of the ideas from Cardello's book---Speed to Spare and I've caught some $20+ horses using this angle.

The second example is tricky. Looks like the 90 is a throwout. But if a 90 Beyer makes this horse a contender, through my personal contender selection process, I'll definitely use this horse at long odds.

[sq764]

In Pace Makes the Race, Brohammer basically says that you should try to use the horse's last effort at all times, unless it's excusable.. And that a horse should show SOMETHING in a bad race, and if he does not, it's cause for concern regarding his form..

I pretty much follow this mindst, although I do allow a little more leeway for a bad race..

[cj]

You are going to be betting a lot of chalk unless you find a way to discover "excusable" races the public can't readily identify.

[sq764]

You are going to be betting a lot of chalk unless you find a way to discover "excusable" races the public can't readily identify.
No way would I play chalk.. I only play 6/1 or better.. I use the last race as much as possible only if it fits the distance/condition of today's race.. That could be 2-3 back..

Playing Penn, CT and Mnr, I get all kinds of longshot opportunities..

[sq764]

Here's a good example from tonight's Penn National card:

│ -----------------------------WELSH SPIRIT-----------------------------------
│ 16Jan05 5 PHA FT 5.0 D CLM 22.1 45.7 58.4 6.0 9.0 4.3 7TH 81 S
│ 14Dec04 10 PHA FT 7.0 D CLM 22.8 46.2 125.8 4.5 8.0 43.3 9TH 25 S
│ 30Nov04 2 PHA FT 6.0 D CLM 22.2 45.2 110.6 1.8 4.3 16.0 9TH 57 P
│ 16Nov04 8 PHA FT 5.5 D ALW 22.3 46.2 105.5 2.0 6.0 18.8 9TH 59 P
│ 1Nov04 8 PHA FT 6.0 D ALW 22.3 45.9 112.5 2.0 5.8 15.8 7TH 48 P
│ 17Oct04 8 PHA YL 5.0 T CLM 23.1 47.5 100.2 1.8 2.0 1.3 2ND 73 P
│ 20Sep04 8 PHA SF 5.0 T SAL 22.2 46.1 58.5 2.5 2.0 0.0 WON 79 P


Tonight's race is 5.5F on the dirt.. What would you make of this horse and which paceline would you focus on?

[JohnGalt1]

I calculate the William L. Scott PCR figure, modifying it a little, which forces me to look at all races in the past performances which I need to do to create it.

Michael Pizzola answers the question of which pace line to use. He says in Handicapping Magic to use the last pace line unless you shouldn't. Then he lists the reasons why. Sometimes it's correct to consider all 10 races.

I use his guidelines. One exception is a layoff horse. I will use races after a layoff unless the race is on a different surface or the horse was never competitive. Then my pace speed fig will be placed in parentheses.

I also combine a two race pace figure into one to hopefully get a more meaningful assessment of a horses ability.

[andicap]

Here's a good example from tonight's Penn National card:

│ -----------------------------WELSH SPIRIT-----------------------------------
│ 16Jan05 5 PHA FT 5.0 D CLM 22.1 45.7 58.4 6.0 9.0 4.3 7TH 81 S
│ 14Dec04 10 PHA FT 7.0 D CLM 22.8 46.2 125.8 4.5 8.0 43.3 9TH 25 S
│ 30Nov04 2 PHA FT 6.0 D CLM 22.2 45.2 110.6 1.8 4.3 16.0 9TH 57 P
│ 16Nov04 8 PHA FT 5.5 D ALW 22.3 46.2 105.5 2.0 6.0 18.8 9TH 59 P
│ 1Nov04 8 PHA FT 6.0 D ALW 22.3 45.9 112.5 2.0 5.8 15.8 7TH 48 P
│ 17Oct04 8 PHA YL 5.0 T CLM 23.1 47.5 100.2 1.8 2.0 1.3 2ND 73 P
│ 20Sep04 8 PHA SF 5.0 T SAL 22.2 46.1 58.5 2.5 2.0 0.0 WON 79 P


Tonight's race is 5.5F on the dirt.. What would you make of this horse and which paceline would you focus on?

For me I'd focus on the Nov 30 or Nov 16 race depending on how the pace of that race fits in with todays. As a horse with high %E, I want to see if there's a chance there's a slower pace today that he can handle and perhaps extend himself further.
I NEVER (well never say never, but rarely) use 5f races as pacelines. They are screwy races and never adjust well to longer dashes. Having said that I will use the races in my analysis of the horse's form cycle. The 81 says to me the horse may bounce off that line since cheaper horses tend to keep their form for much shorter periods of time. (Fillies as well.)

I tend to look for the horse's fastest velocity in his last three races at today's surface (fast/dirt, etc) and within 1 furlong of today's distance. I'll often just take the longest odds horse of the top three unless there's a good reason to dismiss that horse. So I'll start from the top of the odds board and work down. If I don't like a higher odds horse in the top 3, I'll check the 4th best horse to see if he has the odds and positive indications that I look for.
(form cycle, trainer change, improved posts, better pace situation today, energy that fits the race, etc.) However I'll forgive a lot of negative stuff if the horse' has any sort of positive indication in his PP and is high enough odds.

HOWEVER I will tend to downgrade one aberational performance. In order words if the PP that qualified him was substantially better than anything he has done in his last 1 or 2 form cycles I'll be very suspicious and want to know WHY he improved and could he repeat that performance today. Look at all the horses that throw in one great effort because all the stars were aligned that day and he never came back to it. Or that great effort knocked him out. ANyone who read Odds Must be Crazy will remember the Williamstown example -- ran an amazing 1 mile in the Withers at age 3 and never ran that fast again.

[sjk]

Since no one seems to be speaking up in favor of averages I will do so. I am using the average of the last 5 races (all dirt) with an adjustment to keep any outliers from overly affecting the result.

As to Larry's question of how you know what is most predictive my answer is to throw a large amount of data at it and try to find the method that gives the smallest std dev of error.

[JohnGalt1]

Do any of you receive The Handicapper's Edge newsletter from Bris? In the colored insert that comes with some issues from All Ways they list positive factors in certain types of races.

Some races have a positive ROI when averaging the 2 fastest of the last 3 races.

[garyoz]

IMO of what it is worth, after a long detour into handicapping software (I have assembled quite a collection) I have come back to where I started and that is form cycle analysis. I think that computer programs work well with Sartin/PIRCO methodology of paceline selection, the question comes down how well does paceline selection work with picking winners and/or showing positive results. I think the results are mediocre at best if you are comparing one or averaged pacelines for each horse in a race. It is unlikely that each horse will run back to the selected race, also recent past races do not capture moves forward or bounces.

If you are looking for an automated method of paceline selection, I would suggest best 1 of last 3. The most comprehensive and rigorous analysis of speed figures was the Sport Stat study back in 1994 (yes it is old). The study compared about 10 different speed figs and power figures. In terms of the most profitable results for just about every speed figure, the best of last three was most profitable but not necessarily the most winners.

In using paceline selection approaches, I had my greatest success when I ignored conventional wisdom and would go back sometimes more than 6 months or a year and use a paceline that I thought reflected a race or the horse's condition that mirrored the current race. I also had success by selecting races based upon matching the expected pace of the current race with the pace of race for a past race. I also had some success just taking the race with the top Total Pace Rating if that race made some sense.

A couple other comments, moving averages are used in statistics & econometrics in order to smooth curves. You will actually downgrade peak performances and upgrade poor performances using this approach. I think you want to focus on variance instead of trying to adjust it out of the model. Winners are usually putting in peak performances, and you want to bet against favorites in weak performances. What is useful to identify trends in security analsyis is misapplied to horse racing (IMHO).

I think averaged pacelines should be the exception not the rule. Ken Massa at HTR has written an excellent article available for free on his web site on different approaches to paceline selection.

Another comment, I know the Allways program very well. I think that it's automatic paceline selection isn't very good. Statistically averaging the best last 2 of 3 races pacelines outperforms its automatic single paceline selection in Allways, that's why they hype it. They can't go back to improve the algorithms for paceline selection because it would make the databases that use the costly files worthless. They can't make fundamental changes to the program, they can only add to it incrementally. That's why the next version of ALW is supposed to focus on bet structuring (IMHO).

If you are interested in data for form cycle analysis there are CJ's figures, the Xtra's (equiform), Thorograph, The Sheets, On-Target (Trackmaster) as some of the tools.

[sjk]

garyoz,

I can't imagine that anyone had a database of any size whatever in 1994. Now there are many of us who can test ideas on a database with 2-3 million records.

As I said above, I have found an adjusted average to test best and to lead to success. I certainly would not expect horses to run right back to the average value. It is all a matter of probabilities and once you have established pace values for each starter you can begin to look at how that affects the probabilities of each one winning today's race.

[jfdinneen]

Sjk,


I fully agree with the use of average (Median) performance for older horses as you can expect Regression to the Mean to be a significant factor particularly following an exceptionally good or bad performance.

As I have indicated elsewhere, we judge sporting prowess in baseball, for example, by batting averages not by any individual performance so why do we focus so much on best performances in horse racing?

Best wishes,


John

[jfdinneen]

Thoroughbred,


If you are dealing with older horses (4yo+), then Regression to the Mean will be much more significant than any recent best performance and so calculating a Moving Average allows you to include the recent best performance with average ability. Personally, I find a three race moving average very productive.

On the other hand, if you are dealing with younger horses (2yo, 3yo) then I agree you should focus on best performance as these horses are still improving.


Best wishes,


John

[sjk]

Sorry to reply to my own post but I was thinking about garyoz's suggestion that we embrace the variance rather than model it out. At first thought this seems like a good idea. On the other hand I think it depends on where you think the variance comes from.

I would think a major source of variance is measurement error. Of course, you work hard to refine your pace measurements but even so there is bound to be error. Another source of variance is the usual assortment of good breaks, bad breaks, bone-headed tactics, and the variation in how the horse is feeling that day which is common to all horses.

The fact that one horse shows more variance in his past few races than another may or may not be useful data about the horse. As I said above, I have had good success in washing it out.

[garyoz]

The 1994 study was for the Hollywood meet. I am not away of any published study on paceline selection using the two or three million sample size you describe. I'd be interested in any such studies. I have looked at a few thousand race database for Allways and I know their single selection paceline algorithm underperforms.

In terms of the variance issue, I can understand your pointview. I just don't think a horse's probable performance in today's race is best represented by a normal curve and conceptually I don't see performance variance as sample error. I see performance variance as primarily form cycle. Just my opinion, and opinions make markets I certainly wish you best of luck with your approach.

[sjk]

Garyoz,

Where would one publish a study on handicapping? If you have some alternatives you would be interested in seeing compared, I could take a look at it for you. The only problem is that I would be using my definition of the pace figures and that might not line up with your definition.

What part of town are you in anyway? I'm up here in southern Butler Co.

[garyoz]

The Sport Stat was a study that was sold for $45. I would think there is a market for a valid and reliable study on paceline selection. Also the old Meadows Newsletter did alot of really good studies. Won't be something coming out of Bris. A don't know the Sartin literature completely, but there were alot of things published. Personally, I'm really moving away from this methodology. I was just sharing opinions from my experience.

I live down in Hyde Park--drop me a private email and maybe we can meet at River Downs. I don't play the local tracks much--but they have a nice race book.