How To Make A Power Rating

[PriceAnProbability]

Thought this deserved its own thread.

Unlike a speed figure or pace rating, a power rating has to take literally everything into account, because if it doesn't, whichever factor you are overlooking is going to consistently screw up your value line, along with your exotic tickets.

I found that working on a power rating method forced me to consider literally every factor. After several years, I've isolated enough factors that my numbers usually come close to the program morning line, with only a few horses deviating by much (these are the basis for most plays or throwouts).

As I said in another thread, the signs that you're making a good power rating system are that the top rating is profitable as a win bet, and the horses you rate above the public (in rank), the "value" horses, also show a profit, while most of your morning lines are at or near the program (adjusted to a 120-point base for takeout).

Has anyone ever found a power rating that is profitable among the commercial services?

[betovernetcapper]

All of them are are going to be profitable some of time at some tracks in selected distances or conditions. Better programs create power ratings that are sensitive to current track profiles, but profitabilty is a function of odds and win percentage and I find redboarding helpful. :)

[kitts]

There is some software out there that allows the user to choose the factors and weight them according to individual preference. I use All-In-One V6 and I have "customized" my power rating formula. The software generates a Fair Odds line based on the power rating. The top horse at Fair Odds of 2/1 or lower with at least a 5 point gap in the power rating is usually a good bet, although not always a value bet. I agree that Fair Odds of 2/1 makes tote odds of 3/1 an overlay. But not a bet.

[garyoz]

In order to avoid "double counting" a factor (e.g., class and speed being highly correlated and arguably representing the same factor)--a good approach would be to create variables using a statistical approach such as factor analysis. Otherwise what is the justification for putting variables together and weighting them?

[sjk]

PanP,

If you have a rating where the top figure horse shows a flat bet profit you have done some excellent work. I would doubt that any rating that is available broadly at a reasonaly price would do that well for long because it would attract a lot of users who would bet it well below the break even.

[PriceAnProbability]

In order to avoid "double counting" a factor (e.g., class and speed being highly correlated and arguably representing the same factor)--a good approach would be to create variables using a statistical approach such as factor analysis. Otherwise what is the justification for putting variables together and weighting them?

I don't use variables to construct my ratings, but I do something similar.

Generally, if several factors point to the same thing, every one after the first should be reduced in its weight, even if the factors seem to be unrelated. This is a form of regressing to the mean.

That so few people seem to even attempt this says we're a long way to go from the pools being so sharp that one cannot win. I suspect a few do know how to do this and just pray that threads like this would go away.

[Dave Schwartz]

I don't use variables to construct my ratings, but I do something similar.

What do you mean by this statement?

Would you call (say) "Last Race Speed Rating" a variable?


Dave

[highnote]

After several years, I've isolated enough factors that my numbers usually come close to the program morning line....

What is the difference between a factor and a variable?

[PriceAnProbability]

What is the difference between a factor and a variable?

A variable has less room for human interpretation than a "factor."

[betovernetcapper]

A variable has less room for human interpretation than a "factor."

???????????

[njcurveball]

Originally Posted by PriceAnProbability
A variable has less room for human interpretation than a "factor."
This makes perfect sense. Perhaps it is the terminology that is causing confusion??

In the context of this thread, a variable would be something like days away, or jockey win%. They have numbers that are the same to everyone.

A factor would be long layoff or good jockey. Different people have different interpretations for these.

One person might say 90 days is a long layoff, another would say 180. Some would say 15% is good for a jock, others would say 20.

The problem with making a power rating is that every factor isn't german for every race.

Pace in last race may be a major factor for an older maiden claiming race and meaningless in a 2yo race with lots of first time starters.

The post before about tracks, distances, conditions, etc is the most relevant to winning.

We all have a "system" to pick 33% winners if we play every race in the country. That would be simply taking the favorite. The key to any factor is the difference between how often it wins and how often the public THINKS it wins!!!

Jim

[Robert Fischer]

We can get into semantics pretty quickly, but Factors can be a combination of two or more variables. If we are talking about principal component analysis for racing, we might have two variables that both measure speed. To reduce the two variables into one factor we fit a regression line. The variable that would represent that regression line would be our "Factor".

[betovernetcapper]

For me the words factor and variable, within the context of a power rating ,are interchangable and each refer to an ingredient of the power rating. Ingredients might be speed rating last race-best speed rating-days away-trainer-jockey-trainer jockey combo-post position-projected pace first call-second call-final fraction-earnings-pedigree-finish last ect.

[PriceAnProbability]

Originally Posted by PriceAnProbability
The problem with making a power rating is that every factor isn't german for every race.

Since everyone has the same "problem," however, the question is not who solves it, but rather who solves it better than the rest of the crowd.

The one question I don't see answered here is what the power rating is supposed to represent. I also don't see much discussion of how to convert the ratings into a value line.

[cj]

Since everyone has the same "problem," however, the question is not who solves it, but rather who solves it better than the rest of the crowd.

The one question I don't see answered here is what the power rating is supposed to represent. I also don't see much discussion of how to convert the ratings into a value line.

There have been plenty of discussions about converting ratings into odds lines. Try the search function, I'm sure you'll find a few lengthy threads about this very thing.

[PriceAnProbability]

There have been plenty of discussions about converting ratings into odds lines. Try the search function, I'm sure you'll find a few lengthy threads about this very thing.

I was asking about what the ratings represent, i.e., how many points on a power rating equals one length.

My search didn't find much. Perhaps someone has a link?

Do you use a value line from your figures?

[cj]

Here is a post I made a while back that I think relates to this:

http://www.paceadvantage.com/forum/showpost.php?p=277421&postcount=4