HOW TO-POWER RATINGS

[Doug]

If I wanted to make my own power ratings.

Minimum/Maximum factors to use.

Should they be weighted?

Which factors?

Keep the numbers as small as possible so that a 1 or 2 point diff is actually meanifull?

Thanks for any input.

Doug

[superfecta]

Originally posted by Doug
If I wanted to make my own power ratings.

Minimum/Maximum factors to use.

Should they be weighted?

Which factors?

Keep the numbers as small as possible so that a 1 or 2 point diff is actually meanifull?

Thanks for any input.

Doug I would use them as a base to be used as quantifiers,not as a be all end all.Start small,maybe as basic as three factors.First,last race performance.Second,pace style.Third,Par time for this race.Something to compare the first ratings to.Don't try to make one #.You want to have a picture of possibilities,not one perfect horse.Make your scale small enough that a difference of one point is not a make or break deal.
Most of all,make your scale something you can understand easily,be able to spot irregular # because they are wrong,not because they are right.I hope you can get my drift when I say that.Many times the most logical horse is not the best pick because of a favorable rating.Thats why I say use your ratings as a guide,not as a system.
If you can devise a rating scale that is comfortable to you,then you can build upon it ,introduce new factors,and see if it helps on contenders.You can always rework your scale till it gives you a more reliable rating on the races you choose to play.ALwAYS know what a rating was based on,don't incorporate other ratings that you don't know how the # was produced.
But be patient,experience is the skill that can't be given to you,you have to work on it,it takes time,and that is why most horseplayers don't learn it.
I hope this doesn't sound like I'm talking down to you,I don't want it to sound like that.Be comfortable with your own methods and skills.Don't believe that what works for me will work for you.Take what looks good apply it and see if it works for you.You will find your own style after awhile.Hope this helps you find out what you want.

[Tom]

Dick Mitchell's book Thoroughbred Handicapping as an Investment outlined a procedure that commbined speed and class into a power rating. It was fairly easy to do and I used it for a while, until computers and the internet made a lot of other things available to me. When I was only doing a few races a day at only one track, it was pretty much cutting edge technology.

[Derek2U]

for the math freaks among us ... if U have separate IV's & wish
to use them correctly ... multiply them for a total fig ... U can get
a near EXACT estimate of their Over-Lapping Prs through a EZ
chi square test of independence. hehe

[Tom]

Great minds think alike - see my post on another thread about IV's.

[Derek2U]

hey great mind-full 1, let's play some AQU today with analysing
the races. R#1 is confusing (to me) ... and so I think its outcome
mayBE to volatile (to me) ... and so can I play in your OTB acct?
hehe ....

[Doug]

SUPERFECTA,

Thanks,

What I had in mind, in figuring a power rating were the following factors:

Running Style and/or running style and quirin pts.

Trainer Intent. (the way I look at trainer intent)

Class drops (again how I use class drops).

Comments. Only what I think are positive comments.

I do not use a db so I don't have i.v. values. Just looking for a way to score the factors I like best. Like factors that are sometimes hidden or not observed by everyone.

Any suggestions on how to score the above mentioned factors without using a db?

BTW - I got your drift.

I VERY SELDOM feel like I am being taalked down to on this board. I am just looking for input and ideas. Lots of pretty sharp folks on this board.

Anyway, thanks again.

Doug

[superfecta]

Originally posted by Doug
SUPERFECTA,

Thanks,

What I had in mind, in figuring a power rating were the following factors:

Running Style and/or running style and quirin pts.

Trainer Intent. (the way I look at trainer intent)

Class drops (again how I use class drops).

Comments. Only what I think are positive comments.
As long as you know what your scale means(ie.#1 trainer at the meet is = to say,100 pts, second trainer at the meet is 99,ect.)
Quirin pts is rateable,Class could be ,but I don't know how you could quantify trainer intent or running style into a #.

Any suggestions on how to score the above mentioned factors without using a db?Most likely you would want to incorporate a comment line into your rating,only way I can think of to have specific comments on trainer intent or trouble lines for a past race.Good luck with programming that.I wouldn't use a large Db in the first place,so I would use the info in the DRF.The meet standings at the front of the track PPs.That would be current,and would change over time ,so you would have to keep it updated,but not have to maintain a separate large Db.

BTW - I got your drift.

I VERY SELDOM feel like I am being taalked down to on this board. I am just looking for input and ideas. Lots of pretty sharp folks on this board.

Anyway, thanks again.

Doug No problem ,I hope that helps you

[Doug]

SUPERFECTA,

I was reading a book (KINKY HANDICAPPING) and the author suggested " putting a number on your thoughts". I found this to be quite interesting and thats what got me started on this idea.

Maybe a chance to be a little more creative than just looking at computer generated numbers and studying i.v.,etc.

Have to dig in the DRF though which can be quite tedious. Have to cut down on tracks, but maybe thats a good thing.

Thanks,

Doug

[Dave Schwartz]

DOug,

If you're not against creating your own number out of "thin air" (so to speak), try this:

Suppose you want to use 4 factors... Let's call them Speed, Class, Early Speed and Form.

We'll begin by ordering them from most to least important.

Let's say that you think that is:

ES - Form - Class - Speed

Next, assign a weight to each factor. A good place to start is "Fibonacci" where each factor is about .62 rimes the value of the more important factor like...

100 - 62 - 38 - 24

Now create a simple set of "fuzzy grades" like...

Excellent - Good - Average - Fair - Poor

Now, let's see these fuzzy ratings as 1-2-3-4-5


You wind up with a form that looks like this:



100 62 38 24 = 224
Horse ES Form Class Speed
1 3 1 2 2
2 4 1 3 3
3 2 3 3 3
4 1 4 3 4
5 5 4 2 3


Multiply each horse's score by the weight for each factor and divide by the total of the weights.

Thus, the #1 horse gets:
3 x 100 =300
1 x 62 = 62
2 x 38 = 76
2 x 24 = 48
total =486




100 62 38 24 = 224
Horse ES Form Class Speed
1 3 1 2 2 = 486
2 4 1 3 3 = 658
3 2 3 3 3 = 572
4 1 2 3 4 = 434
5 5 4 2 3 = 896


Next, you divide each horse's score by the total of the weights (224).



100 62 38 24 = 224
Horse ES Form Class Speed Score
1 3 1 2 2 = 486 2.17
2 4 1 3 3 = 658 2.94
3 2 3 3 3 = 572 2.55
4 1 2 3 4 = 434 1.94
5 5 4 2 3 = 896 4.00


FInally, convert back to "fuzzy grades" if you want to:



100 62 38 24 = 224
Horse ES Form Class Speed Score Fuzzy
1 3 1 2 2 = 486 2.17 Good
2 4 1 3 3 = 658 2.94 Avg
3 2 3 3 3 = 572 2.55 Avg+
4 1 2 3 4 = 434 1.94 Good+
5 5 4 2 3 = 896 4.00 Fair


Build a little database of your own and decide how "good" horses that go off within a certain odds range do.

Get the idea?

Not complicated. You are in charge instead of software.

The key to making it work is:

1. The criteria you use for assigning the "fuzzy grades."
2. The weights you choose for the factors.
3. The experience you gain with the final grade vs. the odds.

Since your own handicapping is involved, race "types" are not too important, you will quickly find that it does not take too much database to come up with an approach that works. You can even build a set of your own "race shapes" based upon how many goods there are in a given race and how the public has bet them.

And, although it is simple enough to implement with paper and pencil, a spreadsheet would do it very nicely.

What do you think?


Regards,
Dave Schwartz

[Dave Schwartz]

Dang columns did not line up.

[Doug]

DAVE,

I think thats GREAT.

Got the Fabernocci (or whatever the name is) idea. One ? on that. Shouldthe top number always be 100 and then work backwards depending on how manyfactors you use?

The other question I have is regarding the scores.

100 62 38 24 = 224
Horse ES Form Class Speed
1 3 1 2 2
2 4 1 3 3
3 2 3 3 3
4 1 4 3 4
5 5 4 2 3
--------------------------------------------------------------------------------


Multiply each horse's score by the weight for each factor and divide by the total of the weights.

Thus, the #1 horse gets:
3 x 100 =300
1 x 62 = 62
2 x 38 = 76
2 x 24 = 48
total =486




code:--------------------------------------------------------------------------------
100 62 38 24 = 224
Horse ES Form Class Speed
1 3 1 2 2 = 486
2 4 1 3 3 = 658
3 2 3 3 3 = 572
4 1 2 3 4 = 434
5 5 4 2 3 = 896
--------------------------------------------------------------------------------
Opps, just got it. Am I correct in saying that the lower the total score, the better?

Also,how would you make a rough ( close enough for government work) type odds line from these numbers?

This may be just what I am looking for.

If it looks good with pen and paper I'll have to learn how to use a spreadsheet.

Thanks a bunch Dave and BTW could you try to get those darnrows straight next time (HA! HA!)

Doug

[Doug]

Dave,

One other question.

100 62 38 24 = 224
Horse ES Form Class Speed Score Fuzzy
1 3 1 2 2 = 486 2.17 Good
2 4 1 3 3 = 658 2.94 Avg
3 2 3 3 3 = 572 2.55 Avg+
4 1 2 3 4 = 434 1.94 Good+
5 5 4 2 3 = 896 4.00 Fair
[/code


When reverting back to a "fuzzy rating" could you suggest any guidelines for point differentials?

Thanks again,

Doug

[Dave Schwartz]

Doug,

Since all of this is sbujective on the part of the handicapper (you), I have no idea how you'd USE these ratings.

This is simply one of many ways to make a meaningful "high-level" number.

Dave

[Rick]

Dave,

Good idea but why use Fibonacci (Golden Mean) type weightings? Do you have a theoretical reason or is it just something that you've found works well?

Another way to do ratings is by using standard scores. If you're familiar with statistics you can calculate the mean and standard deviation for each variable and combine the "z" values. Since the variables won't be equally good at predicting, you'll have to apply further weighting based either on correlations or trial and error. Of course, if you can do all of that then you can probably try setting up a regression model, but that's another subject altogether and most people who try it won't realize how difficult it is to do correctly.

[Doug]

RICK,

What I am trying to accomplish here is to have a method where I can score/rate the horses based on some criteria (might even call them angles) I have observed over the years.

Would like to keep the method of scoring as simple as possible as it will be done using pencil and paper.

Does not have to be that scientific or technical. Just a fairly good final number and then eyeball my worksheet to get a feel for how that number was arrived at (As Superfecta suggested).

At this point everything would come from the DRF based on my knowledge of interpreting the data. (whether that be good or not so good).

Basically looking for hidden type factors

Dave's idea has given me a good starting point as to a srtucture to make ratings. I will probably will fiddle with the Fibernocci numbers, but it will give me a structure that I can work within.

Your posts are always informative, but when it comes to statistics or higher math it goes over my head.

Any way you could lower the bar and input some fairly basic ideas would be much aprreciated.

Thanks,

Doug

[Derek2U]

Why did you refer to the Fibonacci #s? I'm Very Curious.

[Doug]

DEREK2U,

Keeping in mind that I am not a math guy.

Based on using the 100-62-38-24 I understood that to be .62 of the last number. Assigning the higher number to factor I would think is most important. After entering data based on my thoughts on the factor I would sum the numbers arrived at for each horse and divide that sum by the total of the Fibonocci numbers ( or whatever they may be technically called) to arrive at a final number for each horse.

I understand the structure of the concept (at least I think I do) and will probably adjust (boy, I hate using that word) them to a different % of each other.

If I am off base on this understanding of Fibornocci I would appreciate a message correcting me.

thanks,

Doug

[Rick]

Doug,

Understood. Actually I don't necessarily prefer complicated methods since they don't always work better anyway. Ranking systems like Dave suggested are a good simple way to get a rating. But, you still want some way of verifying that your factors are independent enough. One simple way that I've used is just to see what the ROI is when a horse ranks first in both of two categories. If the factors are too similar, the ROI when both are top ranked will be worse than that for the top ranked horse for either factor taken separately. It's a simple way to see if you're headed down the wrong road and will save you a lot of time.

Don't forget that there are some factors where more or less is not necessarily better. For example, if you include recency you might find that 20-30 days is better than categories both higher and lower than that. And when rating early speed by lengths behind, it might be better to give more points to being behind by a couple of lengths than for leading since it won't be so overbet.

I've developed point count methods in the past just by trial and error on a computer using nothing more than just a simple spreadsheet. Of course, some kind of database accessed by a simple program is much better since you can experiment at redefining your categories. Doing it manually is possible but generally takes too long.

Try not to focus too much on getting a high win % though because doing that usually is detrimental to your ROI. The most profitable methods I've ever played had about 18-27% wins. Trying for over 30% wins seems to be the kiss of death, at least for me.

[Doug]

RICK,

thanks.

Doug

[Tom]

You get post of the month for that one.
Great ideas, there and well well explained.
We do something along those lines in problem solving at work, where we don't always have lot of data available and testing everything would take too long and cost too much. It works.

One quesiton, thought...I always thought "Fibonacci " was a fiber-enriched pasta????

[Dave Schwartz]

Rick,

I have been amazed at how many weighting schemes my AI stuff as developed that wind up haiving a very Fibo-like shape.

I chose Fibonacci as a starting point only because I believe you have to start somewhere and equal weights probably is not it.


Dave

[Dave Schwartz]

Tom,

Actually, it IS a fiber-enriched pasta. Order it with a side of proscuitto and everything should "come out all right."

Dave

[Dave Schwartz]

One more thing about "Power Ratings."

WHat are they really? They are a "confidence number," nothing more.

There is no "verification" to do. The issue becomes "How do I use this "confidence number?" THAT is where the further study comes in.

Dave

[Rick]

Dave,

Very interesting. That's why I always say that having experience with something is the best education of all. I'll have to check some of my good weighting schemes (after normalizing of course) to see whether that holds true for them. It also suggests that anything over, maybe five variables wouldn't add much, which is what I've found in general.

[BMeadow]

As one who has compiled power ratings (Master Win Ratings) daily for the past 16 years--which includes a full year of trial and error development in which I went to the track every single day before I ever sold them--I believe they can be extremely useful depending upon who is doing them and how they are done.

I explain how I do them at my website www.trpublishing.com. Go to Master Win Ratings, then click on Instructional Package which is free. It is an eight-page description of some of the factors I do (and don't) use, and why.

If you know the local horse population, I believe it is better to be subjective than objective. For instance, let's say you use Beyer speed ratings in your power rating formula. A horse ran a giant number last time so he'll be getting a lot of credit for that. However, what if he had an all-time perfect trip, it was a different distance, it was far better than anything else he's ever done, today's pace scenario is against his style, etc.? Now that big Beyer doesn't look so good, does it?

Or let's say you include workout speed/ranking as a factor. Trainer A works everything fast, while Trainer B works everything slow. Or a cheap claimer is worked fast as an "advertising" work to entice a claim. Or one work of 59.1 is all out and the other is under a stranglehold, only they both say 59.1h.

Or let's say you use w-p-s consistency as a factor. What do you do with an allowance shipper from Thistledown with big numbers who now faces classier Gulfstream allowance horses?

Or if you use average earnings per start, and now it's October and the horse won all his money in January and February.

I ran a six-page article, "Doing Your Own Power Ratings," in the November 1998 issue of Meadow's Racing Monthly. I go into a number of methods that players can use to compile their own ratings.

No matter how you do them, the key is to be able to verify their accuracy. And you've got to keep them up to date, noting how your higher-rated horses did compared with your lower-rated horses.

In any event, you can use them only as a guide, since every race is different. Simply blinding betting your top number is never going to give you profits, no matter who does that number--because the more obvious strengths a horse possesses, the lower his odds are going to be.

Ratings are not easy to do. But they are well worth your time, particularly if you review each horse's record individually rather than asking a computer program to do this work. Studying each record will help you know your own local horses.

Good luck.

[Derek2U]

Funny... a thread with a math-science technique like weighting
factors with a Fibonacci sequence and using a subjective ploy
like Power Ratings. Art & Science, Science & Art ... I think that
is the ONLY way to develop a winning system. (Fibo accurately describes phenomena) .... But I am a firm believer in using as much math as needed as long as it does not run counter to what your brain knows as correct after playing so many races.
Take regression anaylsis: Of course its a useful technique;
and of course, you should replicate your final weights.
It's one drawback: It was NOT designed for a game like racing
and so it violates the math behind the technique. Racing is
loaded with "subterfuges" ---- racing to get in shape; racing to
try something new like turf or a route or blinkers or against a
higher class or to satisfy some owner's ego. Now these things
imply that the Horses will either NOT try hard to win or have
zero winning chance. And so, that violates the math assumption
behind R squared: a study of variance implies that the measured
score is a true score--- and its not for reasons I just typed.
It's like doing a correlation between Class & SAT scores and after
finding that a lot of students deliberately Sabotaged their scores.
Or cheated & got higher scores. It would be far better to reduce
a Regression to maybe the top half of the field + the Fav and
ignore any who finished way behind. Just my thought on this
snowy NY day.

[Rick]

Derek,

I have to agree with you that regression has limited use in horse racing. One of the biggest problems in all of statistics is assumptions about the distributions of the variables. Usually the mathematical techniques are based on normal distribution assumptions and when the the data doesn't conform to that the techniques break down.

Outliers can be a really big problem because they get weighted according to the square of their distance. That weighting was convenient in developing a technique that is easily computed but often is not justified in the real world. There are some techniques that use the absolute value of the distance that can be programmed as an algorithm on computers but they're not as widely available.

I've used regression for verification of techniques many times but my methods have generally been developed by trial and error more often. That and a good dose of common sense usually produces better results than blindly believing a regression equation.

I could go on and on about how the real problem is determining a good measure of effectiveness that doesn't require a huge sample of races but that's another subject.

[Doug]

DEREK2U,

You wrote,

It would be far better to reduce
a Regression to maybe the top half of the field + the Fav and
ignore any who finished way behind. Just my thought on this
snowy NY day.

Sounds interesting. Could you dumb that down a little for me.

thanks,

Doug

[Rick]

Doug & Derek,

When I've tried using finish position as the predicted variable I've always gotten better results by treating any finish position worse than 5th as 5th place. These horses that are really far out of it are not pressed by the jockey to do their best. My guess is that they'd rather save them for another race.

[rrbauer]

Hey Everybody:

Keep on Truckin!

This is a good thread!!

[Doug]

RICK,

One of the factors I am using is last race performance. Not just finish position, but taking into account - Key race- Better than looked finish-improvement (especially early speed improvement).

Any comments on this?

Keeping the factors down to 3 or 4 is not the easiest thing to do.

What I have been doing is combining a few of the same type of factors to to arrive at a final rating for that one factor.


Doug

[Rick]

Doug,

I don't use the last race specifically in any way but if you can find anything that indicates improving form that isn't overbet it might help. Being within 3 1/4 lengths at the first call in any of the last two or three races (the more the better) may help. Just don't give extra credit for leading at the first call because that may be overbet. I've also used a combination of finish position and class change before to identify when a horse is likely to improve.