Class rankings

[MJC922]

Originally Posted by classhandicapper

I'm not asking you to divulge too much here but what are you looking at?

1. class labels?
2. earnings?
3. entire record
4. recent race(s)
5. finish and beaten lengths?
6. speed and pace figures?

I am currently building a class performance rating system. At this stage I'm not trying to say who I think the class of the race is. I'm just trying to measure a single performance using factors other than fractions and final time. Then I am comparing my results to high quality speed figures to see how they compare in terms of win% and ROI. I've been pleasantly surprised by my results so far in a limited sample but surprisingly it is doing way better on dirt than turf.

Hi Class, lengths, weight, distance and surface are the only inputs the computer uses. Since we're talking about very few inputs and an output ranking which is winning 30% of every race carded on the continent you might guess there's massive data processing involved in generating a measure of a horse's effort for each of its races followed by selection of specific efforts to maximize predictive power. What has astonished me about this ranking quite frankly is let's say we make the assumption that in terms of being predictive the post time favorite can never be improved upon, which I don't think is an unreasonable statement, we know the favorite wins at a 34 or 35% clip. So if 'who beat who' essentially by itself can be processed by the computer to the point where it wins 30% then the rest of the game, the 'kitchen sink' so to speak is only approx. 5%. I can say then with certainty that we're talking about a game which is LARGELY a single factor game of approx. 85% class. It's not a wonder then with a high takeout it's a real challenge to beat. All of these things we as handicappers agonize over, is it the right distance today, the right surface, the post position, who is the trainer, the jockey, is it a negative class drop, the pedigree, physicality, on and on, these other factors have importance and are decisive from time to time but they're all sitting in this tiny 5% 'bucket'. Being in the handicapping game for a long time I found that to be surprising.

[MJC922]

To clarify:

Lengths, weight, distance, surface are the only inputs being used from previous races. NO inputs whatsoever are being used from today's race. That is, the horse could've never routed before or run on turf before whatever the case if it's ranked 1st it's a 30% win rate. I'm not saying it does just as well in those types of situations but I want to communicate that I'm not selectively rating anything based upon any details about today's race. When we look at BRIS prime power or some such what we're looking at is a composite of a lot of factors specific to today's race. There's a big difference.

[classhandicapper]

I'm doing my own study on the subject but it's narrower in scope. When I have enough data to draw some conclusions I'll post some tidbits.

I'm not still not clear on one thing. You said you are using lengths. But what's at the core of the class rating?

For example: If one horse loses a Grade 1 race by 5 lengths and another loses a maiden race by a neck how do you know the Grade 1 horse is better unless you have some sort of class scale defined.

[Capper Al]

Originally Posted by classhandicapper

I'm doing my own study on the subject but it's narrower in scope. When I have enough data to draw some conclusions I'll post some tidbits.

I'm not still not clear on one thing. You said you are using lengths. But what's at the core of the class rating?

For example: If one horse loses a Grade 1 race by 5 lengths and another loses a maiden race by a neck how do you know the Grade 1 horse is better unless you have some sort of class scale defined.

I'm finally finishing up my class study. My samples found that BRIS does a better job than I do for race ranking. So I'm back fitting BRIS RR into my equations. I let you know how it comes out after testing. It will be a mix of BRIS and my stuff.

[classhandicapper]

Originally Posted by MJC922

What has astonished me about this ranking quite frankly is let's say we make the assumption that in terms of being predictive the post time favorite can never be improved upon, which I don't think is an unreasonable statement, we know the favorite wins at a 34 or 35% clip. So if 'who beat who' essentially by itself can be processed by the computer to the point where it wins 30% then the rest of the game, the 'kitchen sink' so to speak is only approx. 5%. I can say then with certainty that we're talking about a game which is LARGELY a single factor game of approx. 85% class.

Unfortunately, there is another way to look at it.

If you test the top speed figure last race using a high quality set of figures alone (especially for just dirt racing), you'll approximate 30% winners (or at least close) using just that single factor.

You could then conclude that approximately 85% of handicapping is just the speed figure.

This has long been "the" handicapping dilemma for me.

There have been times I was very class oriented and used no pace and final time figures at all. At other times I was very pace and final time figure oriented instead. Each would take me toward different horses, but with about the same long term success rate. When they agreed, they would be MORE predictive, but the average prices would fall sharply.

The biggest problem is that viewing things 2 different ways causes a lot of confusion when they disagree. You find yourself second guessing you "class" selections with slower speed figures (because many run up the track) and second guessing your "speed" horses when they move sharply up in class (because many run up the track).

So it has long been my goal to create a hybrid method that either combined them or used the one that worked best answering today's specific questions.

That's a lot easier said than done.

My trial and error experience has been that combining them doesn't improve the overall predictiveness of either method as a stand alone all that much, but drags down the average prices. You may wind up picking a handful of extra winners (MAYBE), but you sacrifice even more in price because you start approximating the thinking of the public more.

There's an answer in there somewhere. That's what I am working on now because I have the data to actually make tweaks and test.

[MJC922]

Originally Posted by classhandicapper

Unfortunately, there is another way to look at it.

If you test the top speed figure last race using a high quality set of figures alone (especially for just dirt racing), you'll approximate 30% winners (or at least close) using just that single factor.

You could then conclude that approximately 85% of handicapping is just the speed figure.

This has long been "the" handicapping dilemma for me.

There have been times I was very class oriented and used no pace and final time figures at all. At other times I was very pace and final time figure oriented instead. Each would take me toward different horses, but with about the same long term success rate. When they agreed, they would be MORE predictive, but the average prices would fall sharply.

The biggest problem is that viewing things 2 different ways causes a lot of confusion when they disagree. You find yourself second guessing you "class" selections with slower speed figures (because many run up the track) and second guessing your "speed" horses when they move sharply up in class (because many run up the track).

So it has long been my goal to create a hybrid method that either combined them or used the one that worked best answering today's specific questions.

That's a lot easier said than done.

My trial and error experience has been that combining them doesn't improve the overall predictiveness of either method as a stand alone all that much, but drags down the average prices. You may wind up picking a handful of extra winners (MAYBE), but you sacrifice even more in price because you start approximating the thinking of the public more.

There's an answer in there somewhere. That's what I am working on now because I have the data to actually make tweaks and test.

Let's talk about how predictive 'good' speed figures are. They hit upper 20's only with selective application. Take them at face value without looking at numbers specific to today's basic distance and surface they hit approx. 25%. I believe Cardello showed the Beyer's around 28-29 but only if you used the same basic distance and surface. In other words a 'representative' race based upon today's scenario. In the Sport Stat study I think Beyer checked in about 25% top fig taken at face value which would be ignoring today's distance and surface, and someone correct me if I'm wrong Ragozin was about 24%. Time is going to bump into the class often enough, these are races so it's unavoidable. As I said for the rankings I want them to be predictive and decouple class from everything else as much as possible, I don't care about prices at all. The prices come later as all of the other factors are examined. Very much like the Beyer figure, if it moves almost 5 points just by bringing it into the context of todays race distance and surface you can probably see potential in something that starts out 5 points higher.

[MJC922]

Originally Posted by classhandicapper

I'm doing my own study on the subject but it's narrower in scope. When I have enough data to draw some conclusions I'll post some tidbits.

I'm not still not clear on one thing. You said you are using lengths. But what's at the core of the class rating?

For example: If one horse loses a Grade 1 race by 5 lengths and another loses a maiden race by a neck how do you know the Grade 1 horse is better unless you have some sort of class scale defined.

I'd rather not get into that level of detail on the forum. There are books dealing with the subject matter of doing similar things. The implementation is more difficult than the concept but still I like to think I have some unique twists built into the process that perhaps others aren't using.

[classhandicapper]

Originally Posted by MJC922

Let's talk about how predictive 'good' speed figures are. They hit upper 20's only with selective application. Take them at face value without looking at numbers specific to today's basic distance and surface they hit approx. 25%. I believe Cardello showed the Beyer's around 28-29 but only if you used the same basic distance and surface. In other words a 'representative' race based upon today's scenario. In the Sport Stat study I think Beyer checked in about 25% top fig taken at face value which would be ignoring today's distance and surface, and someone correct me if I'm wrong Ragozin was about 24%. Time is going to bump into the class often enough, these are races so it's unavoidable. As I said for the rankings I want them to be predictive and decouple class from everything else as much as possible, I don't care about prices at all. The prices come later as all of the other factors are examined. Very much like the Beyer figure, if it moves almost 5 points just by bringing it into the context of todays race distance and surface you can probably see potential in something that starts out 5 points higher.

Speed figures are more predictive on dirt than turf. That's why I put that little blurb in there. For just dirt races, they get a few percent higher than the overall number.

[Capper Al]

Originally Posted by classhandicapper

Speed figures are more predictive on dirt than turf. That's why I put that little blurb in there. For just dirt races, they get a few percent higher than the overall number.

I believe Beyer was both right and wrong about class being speed. Right because speed does tell class. Wrong because class is not speed and must be measured independent of speed to have any real value to the handicapper.

[MJC922]

Always interesting to see who the computer brings into the fray on class -- this week I'm like who the heck is Holy Boss? Oh a sprinter... quintessential unbiased observer. I may start posting the top 100 all ages ranking in some other forum (not Triple Crown) because the 3 year olds are now catching up, AP broke into the top 10 on the all ages with at least three North American starts lifetime and a start in North America over the past 6 months.

1	American Pharoah
2	Frosted
3	Holy Boss
4	Lady Eli
5	Materiality
6	Lovely Maria
7	Competitive Edge
8	Carpe Diem
9	Divining Rod
10	Barbados
11	I'm a Chatterbox
12	Promise Me Silver
13	Dortmund
14	International Star
15	Birdatthewire
16	El Kabeir
17	Ready for Rye
18	Calculator
19	Her Emmynency
20	Condo Commando
21	Huasca
22	Blofeld
23	Cyclogenisis
24	Donworth
25	Saratoga Heater
26	Hootenanny
27	Stellar Wind
28	Nasa
29	Angela Renee
30	Fame and Power
31	A Lot
32	Bolo
33	Papacoolpapacool
34	One Lucky Dane
35	Eskenformoney
36	Victory Is Sweet
37	Danzig Moon
38	Curalina
39	Far Right
40	Sunset Glow
41	Texas Red
42	Woodwin W
43	Celestine
44	Ocho Ocho Ocho
45	Shook Up
46	London Tower
47	Daredevil
48	Take Charge Brandi
49	Vision Perfect
50	Firing Line
51	Leave the Light On
52	Stanford
53	El Deal
54	Lord Nelson
55	Consumer Credit
56	World Approval
57	Oceanwave
58	Include Betty
59	Souper Colossal
60	Lassofthemohicans
61	Spanish Queen
62	R Sassy Lass
63	Nonna's Boy
64	Irish Jasper
65	Bustin It
66	Wisecracker
67	Prospect Park
68	Bar of Gold
69	A M Milky Way
70	Gimme Da Lute
71	Kentuckian
72	Callback
73	Catalina Red
74	Two Weeks Off
75	Mrs McDougal
76	Leap Year Luck
77	Bold Conquest
78	Frammento
79	Wonder Gal
80	Easy to Say
81	Pohdi Pohdi
82	Blame Jim
83	Recount
84	Yockey's Warrior
85	Pine Needles
86	Divisidero
87	War Story
88	Madefromlucky
89	Ostrolenka
90	Keen Ice
91	West Coast Belle
92	Force the Pass
93	Fantastic Style
94	Danessa Deluxe
95	Dubai Sky
96	Cinco Charlie
97	Hard to Port
98	Miss Temple City
99	Fond of Sarah
100	Classy Class

[classhandicapper]

I like what you are doing, but I still don't see how you can get to that list without creating some kind of class pecking order first using race labels, purses, conditions, or something along those lines.

I am using class labels for testing my automated process. I may play around with purses next just to see if I can simplify the process without losing much accuracy.

In real life gambling I look at the actual horses in the race, but I still have a mental pecking order of the class structure in mind.

[MJC922]

Originally Posted by classhandicapper

I like what you are doing, but I still don't see how you can get to that list without creating some kind of class pecking order first using race labels, purses, conditions, or something along those lines.

I am using class labels for testing my automated process. I may play around with purses next just to see if I can simplify the process without losing much accuracy.

In real life gambling I look at the actual horses in the race, but I still have a mental pecking order of the class structure in mind.

Sounds like good stuff.

The top 100 all ages for this week: http://dropcanvas.com/qepy8

[MJC922]

Well here we are on the last leg, I've enjoyed posting the 3yo rankings through the triple crown chase. Good luck to everyone this weekend.

BTW the latest 'all ages' ranking is on the dropcanvas link noted in the prior post -- moving forward I'll try to refresh it every ten days or so.

1	American Pharoah
2	Frosted
3	Holy Boss
4	Competitive Edge
5	Materiality
6	Lovely Maria
7	Carpe Diem
8	Divining Rod
9	Divisidero
10	Barbados
11	Takeover Target
12	I'm a Chatterbox
13	Dortmund
14	A Lot
15	Lady Eli
16	International Star
17	Birdatthewire
18	Ready for Rye
19	El Kabeir
20	Promise Me Silver
21	Bolo
22	Calculator
23	Spanish Queen
24	Blofeld
25	Huasca
26	Donworth
27	Condo Commando
28	Cyclogenisis
29	Nasa
30	Hootenanny
31	Stellar Wind
32	Fame and Power
33	One Lucky Dane
34	Angela Renee
35	Danzig Moon
36	Eskenformoney
37	Far Right
38	Sunset Glow
39	Texas Red
40	Shook Up
41	London Tower
42	Take Charge Brandi
43	Firing Line
44	Daredevil
45	Celestine
46	Curalina
47	Lord Nelson
48	Vision Perfect
49	El Deal
50	Feathered
51	Stanford
52	World Approval
53	Wisecracker
54	Victory Is Sweet
55	Easy to Say
56	Include Betty
57	Oceanwave
58	Souper Colossal
59	Gimme Da Lute
60	Irish Jasper
61	Prospect Park
62	Face the Music
63	R Sassy Lass
64	Kentuckian
65	Granny's Kitten
66	Nonna's Boy
67	Catalina Red
68	Ostrolenka
69	Callback
70	Consumer Credit
71	Two Weeks Off
72	Papacoolpapacool
73	Sentiero Italia
74	Paid Admission
75	Frammento
76	Bold Conquest
77	Galina Point
78	Night Prowler
79	The Truth Or Else
80	Force the Pass
81	Recount
82	Wonder Gal
83	Pine Needles
84	Quezon
85	Blame Jim
86	War Story
87	Heath
88	Madefromlucky
89	Keen Ice
90	Cinco Charlie
91	Global Positioning
92	Om
93	West Coast Belle
94	Native American
95	Bar of Gold
96	Danessa Deluxe
97	Pohdi Pohdi
98	Hard to Port
99	Fantastic Style
100	Dubai Sky

[classhandicapper]

Are you class figures getting equally good results on turf?

[MJC922]

Originally Posted by classhandicapper

Are you class figures getting equally good results on turf?

To be honest I don't have these current ranking results broken out by surface, I didn't pursue it this time only because I have no good reason to suspect much of a difference. Saying it with pretty high confidence because I have the numbers from which the rankings are derived broken out by surface. In that case the Wpct is 3% lower on Turf vs. Dirt but the I.V. is identical (within .02) So at least to my way of thinking the results are equally as good across surfaces. The I.V. is a decent measure for that IMO because if we look at average field sizes in US racing the dirt averages one less horse so the 3% bump is to be expected.