IMPACT VALUES---ONCE MORE

[formula_2002]

TO ALL;
I fail to understand how Impact Values, can be used in calculating an accurate odds line, since the IV becomes a function of the odds themselves.

For example, a Bris_power ranking of 1 has and impact value of 2.4. But that is just an average of all horses running at all final odds.

When the final odds are in the range;
>0 and <=.45 iv is 1.10
>=.55 and <=1.45 iv is 1.03
>=1.5 and <=2.45 iv is 1.05


I don’t understand how a predictor variable (the IV) can be accurate, when it is impacted by the same thing (final odds) you are attempting to predict.

Can you clear this up for me?

Thanks
Joe M

[GameTheory]

You are not trying to predict the final odds, you are trying to predict the winner of the race.

You calculate the IV without reference to any odds. The derivation of IV's has nothing to do with odds, final or otherwise. Nothing.

If you didn't have access to final odds, could you not make an odds line? You would have nothing to base your predictions on? Of course you would. You would have the impact values.

[rrbauer]

Why am I having this strong sense of deja vu?

[GameTheory]

Originally posted by rrbauer
Why am I having this strong sense of deja vu?

Don't worry, I'm going to shut up on this subject from now on. I can only beat my head against the wall so many times...

[formula_2002]

gt, I think we actually agree.

I posted

"I fail to understand how Impact Values, can be used in calculating an accurate odds line, since the IV becomes a function of the odds themselves."


You posted.
"You are not trying to predict the final odds, you are trying to predict the winner of the race."


At lest one of the handicapping softwares promoted on the Bris web site, uses IV's to calculate an odds line.

I dont see Quirin using IV's to calcualte odds in his book "Winning At the Races". I'm going to make sure of that on Monday, when I contact him.

I dont have Fred Davis's book on IV's but perhaps someone here does and can check if Davis intended to use IV' to calculate odds.

----------------------------------------------------------------------------------



.

[cj]

I have Quirin's books, and he never proposes turning IVs into an odds line. He uses them as a way to eliminate horses from contention, at least the way I read it. He was much more reliant on ROI than any IV when it came to betting a positive stat.

CJ

[formula_2002]

CJ, I agree. Page 3 of Quirins book explains how to calculate and interpert IV's

Nowhere in the book does he attemp to calculate an odds line with IV's

Yet, as I say, there is at leat 1 very popular software package and possibly two that uses the IV to calculate an odds line using IV's.

Because of some of the literature I have read about using IV's to
calculate an odds line, I assumed they got their lead from Quirin.
That appears not to be correct.

[hurrikane]

I agree, can't see how IV could be used to create an effective odds line.
what software are you talking about Joe?

[formula_2002]

hurrikane
I want to touch base with Quirin on Monday about using the IV to calculate an odds line within this software package.

[Rick]

I think Paul Peterson and his "Morning Liner" approach was one thing that caused a lot of people to think that using impact values could be used to calculate a line. Of course, that method is doomed unless all of the factors that you use are totally independent of one another. That's a situation that is never achieved in practice in horse racing. It's probably possible to correct for that problem if you adjust all of the impact values towards 1.0 somehow, but there are better ways of combining variables than using impact values.

[Dave Schwartz]

Formula,

Well, be sure to let us know how "touching base" with Dr. Quirin goes on Monday.

Some of us less-connected individuals have tried years to speak with him. For me it has been 12 years.



Regards,
Dave Schwartz

[formula_2002]

Dave, it's been my experience that if B.Quirin dose not respond to my e-mail by 8:30 am of the following day, I generally will not get a response. Well, Its 9:30 here in NY. And I have nada.

Apparently Mike Nunamaker's book "Moder Impact Values" suggest the use of IV's in making an odds line.

I know that All_Ways uses IV's in formulating an odds line.

If I get the time and inclination I may try to set an odds line with IV"s.. Just to prove it one way or the other to myself.

Right now my odds line is doing well.
Past 30 days saw over 500 picks, returning 1 1/2 % profit.

However the odds line is only based on 6000 picks.

Joe M

[rrbauer]

I dont have Fred Davis's book on IV's but perhaps someone here does and can check if Davis intended to use IV' to calculate odds.



Reply:
Davis used his IV's across a number of attributes to arrive at a probability estimate for each entrant. He normalized the individual probability estimates (to 1) and from those normalized estimates arrived at a "fair" odds value that you could use as a betting benchmark. 1 - Pw = Pl : Pl / Pw = Fair Odds

I encapsulated all of his tables into a computer program (we're talking 1980's timeframe here) and did a workout over part of a Santa Anita meet. The problems as I recall: First, the data entry was very tedious and time consuming. Second, the top-rated horses were almost always "underlays" (what's new!). Most of the "overlays" were horses at very high odds (horses that needed an "IV" with something stronger than a statistic to win!).

His work did yield a couple findings that I was able to use going forward: The dominance of speed types on sloppy/muddy tracks and the value of inside post positions on certain turf courses.

[formula_2002]

rrbauer

Just to clear a point.

Is Davis using the IV figure of factor x or the win % figure of factor x to compute the odds line.

Appears from what you write, he is using the win % of factor x,
which would be conventional.

Thanks
Joe M

[rrbauer]

Just what I said.

He uses the Impact Values from several attributes to arrive at a probability estimate by way of his IV tables. Win percentage is not part of his algorithm.

[formula_2002]

That's how I though people were using IV's to calculate an odds line.

Doesn't make sense to me, but I'll try it on my data base.

Given todays computing ability, its a relative simple matter to calculate the win percentages of each factor, mix and match them and find the combinations that yield the bets profits.

Of course, that leads me to another post. Can two, three or four losing factors, combine to make a (long term) winning play ?.

Thanks
Joe M

[Larry Hamilton]

yes

[rrbauer]

Of course, that leads me to another post. Can two, three or four losing factors, combine to make a (long term) winning play ?.


If "winning play" means dollars, sure....low percentages that are compensated for by high odds can yield a profit. However, in total, if you're chasing a negative sum game (as in slot machines for example) then you need luck on your side to win if you play very long.

[Rick]

Some methods produce a better ROI on their second rated horses because of too much correlation between the factors used. Sometimes you can improve the ROI of these methods by taking the horse with the lowest average earnings of the top two. Win % goes down but ROI goes up.

As to whether losing factors can be combined to produce a profitable method, yes of course they can. They should usually be factors that, by themselves, lose less than average though. Of course, if you can find a factor that loses far more than the average you can use that in a negative way too.

[modred]

Fred Davis used IVs to calculate an odds line.

Small book "Thoroughbred Racing: Probability Computation" (1974)


Basic Rating Procedure: Recency
Class
Consistency
Speed

Each characteristic earns a horse a coded numerical rating - either a 2 , 1 or 0.

A. recency (Days since last Race)

14 or less 2
15 to 31 1
Longer 0

B. Class (if Claiming )
> than $500 than Claining price 2
not more than $500 different 1
< $500 lower than claiming price 0

(remember the date of this booklet is 1974)

C. Consistency

How many races has the horse won it's latest 6 starts?

2 or more wins 2
1 win 1
None 0

D. Speed Rating

Note the higher speed rating of each horse's latest 2 starts.

No more than 2 are higher 2
No more than 3 to 5 are higher 1
More than 5 are higher 0


Basic Rating :Non-Maiden Sprints - Less than a mile - Track Fast Good or Slow.

Code Claiming Rating Allowance Rating

2222 61 56
2221 27 25
2220 9 8

2212 42 34
2211 19 15
2210 6 5

and so on .. there are further modifications for Routes, Maidens and 2 year olds and for Tracks and early speed at the various tracks.

At the end add up all the ratings, divide by the total to get a percentage and convert that to odds.

All this can probably be done with Quirin IVs or freshly done IVs.

I recommend doing it for each track ...should be easier with easy access to data with BRIS and TSN etc.

[Tim]

Impact Values – an attempt at a simple definition

In the last few days I've been reading a lot of posts that talk about "Impact Values". Perhaps a basic definition of impact values in horse racing might help. Impact Values serve as a way to measure the significance of an attribute between winning and losing horses in a sample. They have absolutely nothing to do with odds, only probabilities. You can’t assign impact values. You can only calculate them.

The following is from "Modern Impact Values" by Michael E. Nunamaker.
=======================================
IV: Impact Value. This is the most important measuring tool we have. This is the percentage of the horses in the sample that won divided by the percentage of the starters the horses in the sample represent. For example, in a five-horse field, a single horse represents 20% of the starters. In a ten-horse field, a single horse represents only 10% of the starters. Impact Values cut through this fog, and correctly show that a horse winning a race with ten starters is twice as difficult as a horse winning a race with only five starters. And yet this is done in a way that allows races of vastly different field sizes to be combined and still accurately measure the variable under study. This can be critical in measurements where filed size varies dramatically. If horses from post 1 win 12% of the time, but horses from post 10 only 4% of the time, this does not mean that a horse breaking from post 1 is 3 times as likely to win as a horse breaking from post 10. What it really means is that post 1 is filled in every race, even races with small fields. But post 10 is only filled when there are 10 or more horses in the race. While the percentages make post 10 look 3 times less likely to win, impact values will show the true relationship, which is likely to be far less of an advantage to the pole horse than 3 to 1.

An impact value of 1.0 means the horse will win as often as we would have expected by chance. The lower the number, the worse the horse has done. Conversely, the higher the number the better it has done. For example, an impact value of 0.50 means that horses in that category only win 50% as often as they should if selected at random. An impact value of 2.50 means the horses are winning 150% more often than random chance would dictate
=======================================

By the way, this is the first useful thing I've been able to do with his books, they weren't worth the paper they were printed on.

I’ve found that my attempts to generate impact values using compound (multiple) attributes brought me face to face with the “sample size problem” and didn’t help much. The only compound attribute area that did show good stuff was to combine the “odds” attribute in a sample with the base attribute in order to see how much the public recognizes the base attribute.

[Rick]

Tim,

I can't agree with you that Nunamaker's studies "aren't worth the paper they were printed on". The thing that's really valuable about his studies is that it gets you over the idea that their's some magic single factor that's going to produce winners. But if you look throught the variables that lose less than average, you might be able to find some to combine that would lead to a profitable method.

So many people think that they're using a factor that the other players don't know about when that's not really the case. And, in my opinion, his studies debunk a lot of the mythology about how well some of the pace handicapping methods promoted in books really work. For the most part his studies totally confirm some of the studies of smaller samples I've done on my own.

But people are always angry when someone gives them the bad news, even if it's true. They'd much rather think that buying another book or going to another seminar or practicing at picking the right pace line will make them a winner. Sorry, but you can't buy your way to profits in this game. You have to do the research on your own.

[Derek2U]

I'm no expert on IVs or ODDS but I would like to toss these ideas:
When you have a database of past races and wish to research factors, just remember a few basics. First, the MORE factors you research, the MORE likely you are to get "significant" results. If you do 10 measures & measure each for significance, you increase the Pr that chance -- and chance alone -- will make
a meaningless discovery look important. So, don't go on data
expeditions unless you (a) have a REALLY large sample AND
(b) split-up your sample BEFORE hand into two. Second, like
most findings, if your brain says "Thats Lame", ignore your
findings because it probably isn't real, just freakish outcomes.
A few months ago I read a Nunnamaker's work on IVS and
it was really funny how HARD he tried to find meaningful IVs. A lot of his stuff just revealed a lack of racing experience (to me).
Now this Question: Has anyone used ODDS & CHANGES in Odds
over a horse's last 3 races as a Factor? If the public is so good
in predicting the outcome via ODDS then it seems like this info
is a public record in the DRF and maybe can be predictive.
Honestly, I don't think I'b be capable of weeding through it, but
maybe others could.

[Rick]

Derek,

There was something called POPS (Profitable Odds Patterns, I think) being promoted by Tom Worth and Earl somebody. I think it had to do with lower odds twice in the last 3 races. Someone else probably knows all of the details. I don't know of any evidence that it actually worked though.

By the way, I agree that Nunamaker's studies look like he wasn't very creative about how to use racing information. But, how do you know that he didn't save all of the good stuff for himself and sell the worthless stuff to shmucks like us? Isn't that what everyone does who sells racing information anyway? What would you do?

[Doug]

There was POP&TIP.

Profitable Odds Patterns & Trainer Intent Patterns.

Several different odds patterns to look for and best if used with trainer intent patterns,of which there were several.

Probably more useful back then, which if I remember correctly, was pretty much before similcasting, otb's, whales, etc.

Was by Tom Worth.

Doug

[Derek2U]

Hey I never thought that. Of course, and why shouldnt he?
I'd reveal everything I'd uncovered but thats me .... and Tom
lets cap the AQU races again together this weekend. How about
Sunday?

[formula_2002]

For what it's worth..............

Example: Bris Power ranking of 1 has an IV of 2.44. Excellent!!!

It's ratio of actual winners to expected winners is .87

That simply means that if you bet on every Ranking of 1, you will lose 13% on every dollar.

All the milk is taken out of this cow....

You can combine is with anything you like and you will not statistically improve upon the .87 ... any more than a factor that has an IV of .5 with an A/E of .87.

Based on my work, there is no single published factor whose IV value creates profits.
The is no single published factor that creates profit.

I have never seen a study that statistically validates a profitable system over a wide range of various odds using single or a combination of factors.

Go to page 13 of Quirin’s "Winning At The Races" to see the worst example of testing for significance. He is testing on average odds.
Bad Idea..Fabricand tests on each incremental odds range. Good idea.

I can run off 10 systems a day showing significance on average odds... I've been working 33 years to do it on incremental odds.


If anyone wants a excel report on all the bris power rankings, win %, a/e ratios etc. request it of me via conventional e-mail.

Joe M

[Aussieplayer]

I can run off 10 systems a day showing significance on average odds... I've been working 33 years to do it on incremental odds.

Joe,

Please excuse my simplicity....but are you saying that the goal (should) be to be profitable with a given system over all odds ranges? That a system should be "significant" (profitable) over each odds increment?

Although everyone would agree that this is a good goal, you seem to be suggesting that a "system" or "method" that has odds ranges as a requirement are not as, well, valid? Is that what you're in fact saying?

Cheers
AP

[Derek2U]

Guys get real ... I think that a Horse That WINS & Pays like
Under 7:2 is one kinda horse; from $9 to maybe $12 is another;
and so on, so that, to be useful, a system should be based on
what the winner pays, which is sorta the opposite of Value.
Maybe not opposite but different from Value.
I think you'd have to put the field through 3 or 4 systems & see
what horse resembles what kinda PayoFF Winner. I'm not gonna
do that but in a way, its a statistic that will tell U where U should
belong. hehe I'm a Very BiG LongShot.

[formula_2002]

AP
I'm saying that a system SHOULD be analyzed on each odds range, and not on the total average of all selections.

To me the key is looking for a high ratio of acutal winners to expected winners at each odds range.

I have an "inter-active" excel table posted on my web page.

You enter the number of horses, the win% and the odds , and it will tells you if the results are significant.

I work to that criteria.

Take a look at the table and put in some figures..

[Tim]

Rick,

Perhaps I was too harsh. At the time, in 1995, I bought 2 of his books, “Modern Impact Values” and “Longshots – Unabridged Edition”. While I didn’t expect something as good as Asch & Quandt’s “Racetrack Betting” or Lou Halloway’s “The Talking Tote” I did expect to have my horseracing horizons expanded. I opened them with much anticipation. Unfortunately it only took about 20 minutes to realize he was using weak science. He apparently had an aversion to loose ends. Things like combining first time starters with horses that have not raced in a 100 days for “Recency” is sloppy. His criteria for not including a race in his study left much to be desired. Also, his heavy dependence on BRIS numbers was not something to inspire confidence. It has always been clear to me that BRIS has had quality control issues with their generated numbers. At any rate, I was disappointed to find 700 pages filled with flavors of the same one note report. I would have preferred a 20 page monograph that would have adequately covered everything he had to say.

By the way, I don’t add books to my library looking for a “magic bullet” but to learn and to see what is being disseminated to the public. After floundering for a while in the early 80’s when the ROI from betting speed fell, I learned my lesson, I’ve made it a point to try to understand what the current betting fads are. What’s profitable today is sure to be tomorrows losing method. Fortunately as the public money chases the latest fad it creates an inefficient market elsewhere.

Sorry, I must humbly stand by my statement “aren't worth the paper they were printed on” not because they cost me about $200 but because they wasted my time with an inaccurate and biased study.

[rrbauer]

2002 wrote:
Example: Bris Power ranking of 1 has an IV of 2.44. Excellent!!!


Comment:
Can you identify the numerator/denominator values that produce the 2.44 result?

2002:
It's ratio of actual winners to expected winners is .87


Comment:
Again, what are "actual" and "expected" winners? Normally, a ratio as such, would simply state a result in terms of how many were won in relation to how many you expected to win based upon some attribute.

2002:
That simply means that if you bet on every Ranking of 1, you will lose 13% on every dollar.


Comment:
Where did the dollars come from? This was about how many winners and not about how much money. Is .87 the ROI ?

[GameTheory]

Originally posted by formula_2002
AP
To me the key is looking for a high ratio of acutal winners to expected winners at each odds range.


That is certainly desirable, but it is very tough to find such situations, as you have found out.

It is much easier to find factors (or combinations of) that are profitable in SOME odds range, and then only bet them when they are in that range.

If you make a full odds line, you can just make a blended line like we were talking about in another thread, which will remove many of the "false overlays" you had according to your own line.

Formula, do you make full odds lines or do you just set a minimum odds for each pick? If the former, you might try blending the lines and see what happens...

[formula_2002]

RR

"Can you identify the numerator/denominator values that produce the 2.44 result?"


bp=1 horses
4064 horses won, in 13106 races (4064/13106=.31)

all horses
13182 horses won in 104052 races (13182/104052)=.1266)

.31/.1266=2.44

------------------------------------------------------------------------
if there was no take-out actual winners /expected winner =1

If the take out is say 15% the a/e=.85
or a loss of 15 cent sper dollar

For bris power=1 actual winner =4064
The expected winners is the sum of all the various 1/(odds +1)
which equals 4717.

4064/4717 = .86

[formula_2002]

GT
One of my systems ;
2371 plays
appears to be significant at 2-1
4-1
5-1
6-1
7-1

while >0.and.<2 is profitable it lacks significance, as of now.

3-1 is break even. need more race there.
8-1 to 20-1, profitable in 9 of 12 odd sets. need more races.
>20 I dont even count.

of course the best test is on the test 2371 plays...
Most of the 2371 plays are back fitted.

Joe M

[tanda]

I am not sure why Formula considers a 2.44 IV to be excellent.

That tranlates to a 30% win percentage ... exactly what BRIS claims for its #1 (at least on dirt).

My experience is that any reasonable ability rating can pick 30% winners for its top rated ... just as Prime Power does.

Prime Power is a good rating ... but like all ratings, it is how you use it that determines your success.

[Aussieplayer]

>20 I dont even count.

formula,

So, you DON'T actually assert that a "legitimate" system needs to be significant across the whole odds spectrum??
I thought you were saying it does.

Whilst idyllic, I no longer think that a "legit," "vaild," or "significant" system DOES have to be significant across the spectrum.

GT put it nicely: there are simply certain combos of factors (sometimes) that allow you to be profitable, most often only in a given odds range. Sometimes across the spread!!

Cheers
AP

[rrbauer]

"Can you identify the numerator/denominator values that produce the 2.44 result?"


bp=1 horses
4064 horses won, in 13106 races (4064/13106=.31)

all horses
13182 horses won in 104052 races (13182/104052)=.1266)

.31/.1266=2.44

Comment:

I don't understand the "all horses" categorization and the 104052 races. Races that do not have a (bp = 1) horse in them should have nothing to do with this exercise.

What percent of the starters have the (bp = 1) attribute? (??)

What percent of the winners have the (bp = 1) attribute? (.31)

.31 / (% starters with the attribute) = I. V.

With regard to odds' stratification, in this instance, I would try to do a breakout of those situations where the odds are > 2.20 to 1 for the members of the (bp = 1) set. 2.20 is the approximate break-even win rate (.69 / .31). As someone mentioned there may be other factors within a range that will futher improve the results.

Finally, as a matter of curiousity, do you know what percent of the (bp = 1) set are favorites? I'm guessing it's pretty high.

[formula_2002]

Tanda

"I am not sure why Formula considers a 2.44 IV to be excellent"

My 2.44 commnet was tounge in cheek. The point was, it still yileds a 15% dollar loss, as do some iv's at .51 iv's."

IV's can tell you where a higher % of ALL winner are. Unfortunately, they yield $ losses just as do All other horses.
And the losses are based on the final odds, with short priced horses lossing less then long priced horses.
========================================
rrbauer

iv= # winner with factor/# starters with factor
divided by
All winners/all starters.
=========================================
ap
"So, you DON'T actually assert that a "legitimate" system needs to be significant across the whole odds spectrum??
I thought you were saying it does."

For a fact I don't truly know. However, the more odds ranges I find significant, the happier (more confident, and that is what this ia all about) I get .



Joe M

[hurrikane]

formula,

iv= # winner with factor/# starters with factor
divided by
All winners/all starters.

I am probably wrong, I don't lay awake at night caressing IV but I thought IV would be

(# winners with factor/All winners)
divided by
(# starters with factor/All starters)

it doesn't really effect your numbers and I am probably wrong.
IV tests the strenght of a factor, not it's profitablity.

You can't compare apples and oranges.

[formula_2002]

hurrikane

Quirin, explains IV's as

"i.v. statistics are calculated by divinding the the % winners with a given charcteristics by the % starters with that characteristics."

You are correct. I did the math accordingly but described it incorrectly.

Thanks

Joe M

[formula_2002]

hurrikane

you state the iv calculation correctly.

Quirin, explains IV's as

"i.v. statistics are calculated by divinding the the % winners with a given charcteristics by the % starters with that characteristics."



the iv calculation is;

bp=1 horses won 4064
all horses won 13182
4064/13182 = .308%

bp=1 starters = 13106
all horse starters= 104052
13106/104052= .1259%

.308/.1259= 2.44


Thanks

Joe M

[Larry Hamilton]

Let me save youze guys a little time. You may have noticed that the denominator in this problem is .12 plus or minus a little. This is value of a RANDOM occurence! Frankly, it is safe to just use .1225 as the denominotor and in most cases you will be right. Yes, it may be a little off, but if you aren't careful in this game, you may find you are measuring with a micrometer what was marked with a grease pencil and cut with a chain saw.

[rrbauer]

I left this one until now. At 2AM, bed seemed like a better option!

2002 stated:
if there was no take-out actual winners /expected winner =1

If the take out is say 15% the a/e=.85
or a loss of 15 cent sper dollar

For bris power=1 actual winner =4064
The expected winners is the sum of all the various 1/(odds +1)
which equals 4717.

4064/4717 = .86

Let's take this one:
"The expected winners is the sum of all the various 1/(odds +1)"

Comment:
For one race, the sum of the various 1/(odds + 1) will simply approximate 1 + TR , where TR = takeout rate (with a small distortion due to breakage). For more than one race, we can state that as, (1 + TR)r1 + (1 + TR)r2 + .... (1 + TR)rN. (Where N = number of races being evaluated.) And if TR is a constant then it's simply N x ( 1 + TR).

So, if TR = .18 then those sums (in this case) will simply be 4064 x 1.18; or, 4795.5.

None of that has anything to do with "expected winners" in any application that I'm familar with. "Expected winners" would be derived using the .31 rate that the (bp = 1) set earned in the sample described, as a probability estimate against fresh data to provide a benchmark to the actual win results from that data.

You could extend that to include dollar yields from betting which become "expected value" benchmarks, although in horse racing literature, $ROI seems to be a more popular measure.

If I were to limit the "sum of all the various 1/(odds +1)" to just (bp = 1) winners (4064) then the sum of those odds' reciprocals would be considerably smaller than the 4717 that you derived. Indeed, it would be less than 4064.

It could be that I'm misinterpreting something that you've stated.

[formula_2002]

RR

"If I were to limit the "sum of all the various 1/(odds +1)" to just (bp = 1) winners (4064) then the sum of those odds' reciprocals would be considerably smaller than the 4717 that you derived. Indeed, it would be less than 4064"

By actual calculation, the sum of all the 1/(odds+1) = 4747.

If It were less then 4064, we could all make a profit.
That can best be explained by looking at a dutch bet.
Where a ducth bet is 1/(odds+1) x a fixed unit.

Joe M

[rrbauer]

2002 wrote:

By actual calculation, the sum of all the 1/(odds+1) = 4747.



Comment:

Which is meaningless from an expectation perspective in the example that you're using.

[formula_2002]

RR

if you willl me allow me dutch (1/(odds=1) every BP=1 horse , which according to you would be less then 4064 and you will pay me for all 4064 winning horses, I'l will make you my bookie.

Joe M

I'm sure we are not talking about the same thing.

[Rick]

I'm confused. If I was testing a factor or method, I wouldn't be looking at impact values at all. I'd be trying to find out if the selections had an expected value significantly greater than break-even or some minimum positive value. It's not that hard to do if you just calculate the mean and standard deviation of the payoffs. You can do that on most calculators these days.

[Tom]

Originally posted by Larry Hamilton

....but if you aren't careful in this game, you may find you are measuring with a micrometer what was marked with a grease pencil and cut with a chain saw.

When did you visit my factory? I missed you! <G>

[formula_2002]

Rick

"I'm confused. If I was testing a factor or method, I wouldn't be looking at impact values at all. "

Neither would I but some software programs.

Joe M

[GameTheory]

Originally posted by Rick
I'm confused. If I was testing a factor or method, I wouldn't be looking at impact values at all. I'd be trying to find out if the selections had an expected value significantly greater than break-even or some minimum positive value. It's not that hard to do if you just calculate the mean and standard deviation of the payoffs. You can do that on most calculators these days.

Rick,

What do you do exactly with the mean & sd once you calculate them?

[Rick]

GT,

Here's an example from 620 races from last year that I calculated those values on.

N =620
M = 2.48451612 (Mean payoff)
S = 5.94071612 (Standard Deviation of payoffs)

SE = Standard Error of Mean = S / Sqrt(N) = .2385849

That's 2.03 (.48451612 / .2385849) standard deviations above the break-even point (2.00), which is at the 97.88% confidence level. So, there's only a 2.12% probability that this could have occurred by chance. Normally anything at the 95% confidence level or better is considered significant.

This would also allow me to estimate how my annual income might vary based on the expected number of plays. If I assume about 1800 plays per year and do the same calculation, I get a standard error of .1400224 (5.94071612 / Sqrt(1800)). So, the annual mean payoff should have a standard error of about 0.14 which means the mean payoff should be within:

2.3445 to 2.6245 about 68% of the time and,
2.2045 to 2.7645 about 95% of the time.

Translating those into ROI's we get:

.17225 to .31225 68% of the time and,
.10225 to .38225 95% of the time.

Assuming $100 bets that would give us an annual income of:

31005 to 56205 68% of the time and,
18405 to 68850 95% of the time.

As you can see, even with a very profitable method and a large number of plays, the annual income can vary drastically.

Note: The usual practive of calculating limits based on win % and a constant payoff is WRONG even though it's been repeated many times by various authors. Not taking into account the variance in payoffs is ignoring the most important factor.

[rrbauer]

Rick-
Good summary. Is there any significant difference based on racing circuits? I believe that you have indicated that you only play a few circuits. Was that decision driven by +/- experience or by something else?

I just realized (I think) that you're the Rick who used to be in Arizona....is that right? When did you relocate to Reno? (If I have the wrong guy, pls excuse!)

[Rick]

Richard,

I seem to get more plays at some tracks than others and that affects both my overhead costs and profitability. Also, some tracks tend to have higher win % and lower ROI while others have lower win % and higher ROI. It's hard to say which is the best of those alternatives since higher win % and less variable prices allows you to bet a higher % of bankroll.

It's always tough to quote perfomance statistics because they're different for different tracks and also vary somewhat due to seasonal effects. I'm sure there are some tracks that I would do positively miserable at unless I changed the weighting of my factors. My favorite tracks right now are very "average" since my method was developed using an average of several tracks. Also, average tracks don't tend to change as much as those that have a history of a severe bias. I think the trend these days is for track management to try to eliminate any noticable bias if they can. It's just my opinion, but I think the days of "early speed is all you need" are over. It's too bad because I've made a lot of money in the past just playing early speed biased tracks.

I chose tracks that run at convenient times (for me morning is better than afternoon or evening), having relatively long seasons, and between them covering almost the entire year with two tracks running at any given time. Since I'm otherwise retired, I don't want it to be too much of a job and interfere with my other recreational pursuits.

Yes, I'm the same guy who lived in Arizona. I moved to Reno last October. I've lived here twice before so I'm quite familiar with the area and have a daughter living up here.