Form Multipliers for Own Ratings

[Surround]

I was wondering what thoughts some of you have regarding the value of multipliers (or additions) that are applied to different form factors.
What I'm referring to here is factors applied to say Win - Place - Days since last run - Class shift - Track and Distance record etc. etc.
I recall quite some time ago back when I used to buy Prac Punting magazine they had an article on exactly what I am about here but for the life of me I can't find it anywhere.
I gave up buying PP years ago when I woke up to myself.
They had a name for those factors too but my old brain hasn't retained it.

[lomaca]

I was wondering what thoughts some of you have regarding the value of multipliers (or additions) that are applied to different form factors.
What I'm referring to here is factors applied to say Win - Place - Days since last run - Class shift - Track and Distance record etc. etc.
I recall quite some time ago back when I used to buy Prac Punting magazine they had an article on exactly what I am about here but for the life of me I can't find it anywhere.
I gave up buying PP years ago when I woke up to myself.
They had a name for those factors too but my old brain hasn't retained

it.The complete answer to your question is that they are all interrelated, ie. you can have a 100% win and place S/R for a nag that never raced outside of a bush picnic meeting, and as soon as it is racing in a classier company it fails.

To give you a simple answer, that actually works in a way, do your selections according to each criteria separately and see how they perform.

let's say win rating is twice as good as any other, then its multiplier is obvious.
Try it.
Good luck

[Surround]

Hi Iomaca
yes, you're right, and that is where the class factor would apply.
But Class is a difficult factor to apply when you are applying it in a spreadsheet situation as I'm trying to do.
Moving from a 3 length win in a $5000 bush race to running in a $20,000 city race isn't easy to apply factors to (for me anyway).

[AngryPixie]

They had a name for those factors too but my old brain hasn't retained it.Were they "likelihood ratio's"?

http://books.google.com.au/books?id=JNk9AAAAIAAJ&pg=PA23&lpg=PA23&dq=likelihood+ratios+punting&source=bl&ots=R1Xm5FOq0p&sig=TaUz4CmU3cb78dyAqVVksFpJdnM&hl=en&ei=pxNoS4u9Fs-GkAW9npjLCQ&sa=X&oi=book_result&ct=result&resnum=1&ved=0CAcQ6AEwAA

[Chrome Prince]

You are looking for impact values.


What is an Impact Value?

An impact value is an index which is calculated by looking at the percentage of winners that possess a particular characteristic when compared to the percentage of winners from the group as a whole. This is the generally agreed definition.

To make this clearer let us look at the following example:

Type of race/Location: Open class handicaps in Sydney and Melbourne (excluding Welters and Highweights)

Distance range: 1200m to 1399m

Type of horse: male gallopers, 4yo to 6yo

Variable: Finish position at last start.

In a 10 year time span there were 2967 runners that met the above conditions, and from these there were 259 winners, producing an 8.7% win rate.

Posn wins runs win% Ival

1 64 542 11.8% 1.35

2-3 72 687 10.5% 1.20

4-9 97 1264 7.7% 0.88

>9 26 474 5.5% 0.63

total 259 2967 8.7%

Again, an impact value is calculated by dividing the percentage of winners with a particular characteristic by the percentage of starters with that characteristic.

Let us assume that the characteristic in which we are interested from the above groups is that of last start winners.

We can see that there were 64 horses that had finished 1st at their last start that that won, out of a total of 259 winners from all of the groups. So the percentage of all the winners with that particular characteristic is 64/259 = 24.7%.

Next we find what percentage of the total runners had that characteristic. Here we can see that of the total number of 2967 horses there were 542 that had finished 1st last start. Therefore, the percentage of starters with that characteristic is 542/2967 = 18.3%.

Finally, to calculate the impact value you divide the percentage of winners with a particular characteristic, which is 24.7% for last start winners, by the percentage this sub-group formed as part of the group as a whole, here the last start winners make up 18.3% of the total number of runners.

Thus … 24.7% / 18.3% = 1.35 = the impact value

An impact value of 1.0 means that the horses with a particular characteristic won a percentage of their races that was equal to the percentage they were of the total number of runners in the group. For example, they won 16% of their races and they made up 16% of the total number of runners. Thus 16%/16% = 1, an impact value of one.Thus we can say that horses in this sub-group won exactly their "fair share" of the races won by runners in the groups as a whole.

As the worked example above showed an impact value of 1.35 we can interpret it as telling us that as last start winners this particular group of 4yo-6yo male gallopers, when competing in this type of open class races in Sydney, has historically won 1.35 times their fair share of races, compared to other 4yo-6yo male gallopers on the Sydney scene.

Compare this to those horses in the above table that finished 10th or worse last start, the latter group winning only 0.63 of their share of the races.

This is a useful way to view the significance of the impact value. It shows whether horses possessing a specific characteristic (eg back in 7 days or less, last start winner, beaten 3.1 to 5 lengths last start etc) have won more than their fair share of the races when compared to the group as a whole.

I readily concede that the actual example chosen to illustrate the meaning of the impact value sheds no new light on our understanding of the importance of good recent form. We all "know" that last start winners are likely to outperform horses that were 10th or worse last start. However, when it comes to days since last run etc, the impact values provide an insight that is not normally available.

In the normal course of processing, my computers calculate impact values for over 100 different variables. This mass of data is really overwhelming. Therefore, I have developed what I call composite impact values and that is what I am publishing in the Wizard.

For example, rather than listing various impact values for (say) days since last run, runs from a spell, beaten margin last start etc etc, I combine a number of the individual impact values that I have found relate directly to (say) fitness and use this composite impact value to report on the significance of the fitness profile of each horse.

In the [edited] three major factors are currently covered by impact values… fitness, form, and distance. At this stage I don’t plan to add more factors, just concentrate on these fundamentals. There is time to add more later.

In producing the [edited] impact values I have grouped horses by age and sex in the following way:

2yo colts and geldings

2yo fillies

3yo colts and geldings

3yo fillies

4yo, 5yo, 6yo horses and geldings

4yo, 5yo, 6yo mares

7yo and older horses and geldings

7yo and older mares

So when you see an impact value against a horse you know that this is the impact value that applies to horses of that particular age and sex, recorded in the class of race being contested tomorrow, over a distance within 10% of the distance of tomorrow’s race, and in the same location.

(Note: Location is determined by the way I have broken Australian racing into various levels of competitiveness for purposes of analysis and comparison.)

For example, if we had a 3yo colt (or gelding) contesting a welter handicap over 1600m at Randwick, the Wizard impact value would show how, over the past 10 years, 3yo colts and geldings have performed in 1600m (+/- 10%) welters run on all Sydney metropolitan tracks when possessing the same fitness, form, and distance ability characteristics (profile) possessed by this 3yo runner.

Taking this explanation a step further, if the 3yo was having his 4th run from a spell, was back in 21 days since his last run, was beaten less than 3 lengths at his last start etc, it is this fitness profile that is being reflected in his [edited] fitness impact value.

If the [edited] impact value in this case was 1.40 we can say that the group of 3yo colts and geldings (of which this horse is a member) possessing this particular fitness profile has performed very successfully, winning 1.4 times its fair share of this type of race when compared to all 3yo’s which have similarly competed.

So with the [edited] impact value you are able to see at a glance how certain age/sex groups have previously performed in the forthcoming class/distance/location. Before now this has never been possible. The Wizard now makes the impossible, possible.

This has been a necessarily short introduction to the [edited] impact values.

Warren Block, publisher

Infamously another site used impact values and rated a heavy odds on chance at 500/1. The horse won easily. :D

It is not the value itself which is easy, but how it is effectively used when combined with factors.
Knowing what to combine and what not to is more important than any one impact value.
Horseracing is far harder to evaluated than soccer or tennis or any other sport, because there are hundreds of extra variables.

[beton]

The education continues. i strongly suggest that one uses the search function and types in impact values. Some nice relaxing reading.
Regards Beton

[Surround]

Thanks CP
Impact Values is the term I was looking for.
I'll do some more research as Beton suggests.

Thanks also AngryP
I'll have a read of that and see if it gives me something.

Back to trying to put a figure on "Class", I thought Shaun made a pretty good fist of it in his early "New Ratings" spreadsheet where he used prize money change and beaten margins to give a class figure.

[beton]

more on the subject
http://www.phahorseracing.com/study.html
Regards Beton

[Surround]

Thanks beton
Just some of the stuff I was looking for.
Now all we've gotta do is back the nag dropping two classes and was in the money last start then order the Rolls.

[lomaca]

Thanks beton
Just some of the stuff I was looking for.
Now all we've gotta do is back the nag dropping two classes and was in the money last start then order the Rolls.Now if we can come down to earth and reality, why don't you just allocate points to each class in descending order according to the OZ race classes?

Like, there are say 60 classes (I don't know) the top is 60 point the bottom is 1?

This way you can say that 3 len. win in class 60 worth 180 points for example, while the same 3 len. in class 10 worth 30? (very rough calculations but realistic)

Later on you can refine it, including the price money, location etc.

[Chrome Prince]

more on the subject
http://www.phahorseracing.com/study.html
Regards Beton

That site has good examples of how to calculate impact values, but they fall way short in drawing conclusions based on too small a sample size.
5,000 races is a drop in the ocean, I wouldn't be happy with 50,000.
The guys such as the Reads and Bartholemews have sample sizes in the millions.

The reason for such sample sizes is because not all horses handle factors the same way.
For example, some horses carry weight better than others, so if the impact value of topweights were 1.10, you are over crediting or under calculating for some horses than others, so the prices or ratings will be quite out of whack.

Another example is last start winners.
if the impact value is 1.10 for won last time out, and .90 for didn't win last time out, then a last start group1 winner is going to be under credited and a last start hanging rock winner over credited.

There are so many variables, that drilling down within the data is the only way to go, which is why you need huge sample sizes.

Perhaps this is something I might have a crack at in the future, but it's a mammoth undertaking.

[Surround]

Maybe so that it is a relatively small sample, but it is a "start" point.
The problem after that is to get enough data analysis carried out to be more confident in the IV figures.

Iomaca
A reasonable suggestion to start with. Working out how to extract the different classes from the spreadsheet is the biggest hurdle to overcome.
For instance : SC1MW - BM75 - RB71 - Open etc.
Might be able to do it with a Lookup function but I don't know.
Back to the grindstone I reckon.

[Chrome Prince]

I would think that there is probably a better way than using all the different States classes. It is a nightmare trying to line them all up, and then we have Opens which aren't really Opens etc etc.

I should have something to contribute forthwith ;)

[beton]

Chrome Prince
Any input from you is welcome. The problem is, going down this track, in sample size, (I have 32000 races 27 months) and once you have sufficient races to test then there is the chore of analysing all the data. As you say a mammoth task. The question then bears asking "How relevant is it to the next race?" The Eagles won the premiership twice in the early 90s. Could they win again on those results playing the same winning way? Extremely unlikely. The game has evolved and is 12% faster today. From what I am seeing racing is an evolving sport as well. The trainers are learning from the past and applying it to the present. They represent the business side of racing and have a vested interest. They have to be successful, it affects the bottom line. Punting however is a negative sum aside to racing. We fight for an edge over a smaller pot than we collectively contributed and we endeavour to get more out of that smaller pot than our co-contributors. We must then use old trends to recognize new trends. an example is days to last start. With todays training trends <21 days could easily be put out to 35 days. Sorting out class and how the class relates to races is a major edge. I know nothing worthwhile about class but some of these horses seem to be rising or dropping 3 to 4 classes.

So the bottom line is "that we have to start somewhere and talking gets us moving forward". I know that it gets hard for senior posters dealing with newbies but we do bring a different prespective. Many a gold mine has been found by looking at the problem differently. Regards Beton

[Chrome Prince]

One way of measuring class is the first 600m time, the last 600m time, the overall race time.

Assuming we stick to 1200m races for the moment for ease of comparison.
What is a better guide race time, pace or finishing speed?

Here are some impact values for horses that won over 1200m within 14 days.

**This is a guide only, you need a lot more data to accurately predict the impact values***

3,248 horse won over 1200m within 14 days
692 won their next start.
21.31% S/R 8.44% loss on turnover at tote prices.

94 horses ran their 1200m in under 69 seconds
26 of them won next start
27.66% S/R

52 horse ran their first 600m in less than 34 seconds
11 of them won their next start.
21.15%

214 horses ran their last 600m in less than 34 seconds
53 of them won their next start
24.77%

The actual race time and last 600m appear to be better guides than pace.

So we combine a fast race time with last 600m fast time (now it's been said that a fast last 600m will result in a fast race time anyway, I disagree, it depends on the pace.)

And here is the proof.

Combine fast last 600m with fast race time and we get
27 qualifiers
7 winners
25.92% S/R

Combine fast race time with fast early 600m and we get
19 qualifiers
3 winners
15.79%

Of course these are tiny sample sizes, but hypothetically these should be the best of the best, and they aren't!
So far the race time is the best indication regardless of pace or last 600m.

So what if we reverse what we believed to be true and look at it mirrored.

3076 horse ran their race in over 69 seconds
649 of them won next start
21.10%

3116 ran their first 600m in greater than 34 seconds
665 of them won next start
21.34%

2958 ran their last 600m in greater than 34 seconds
624 won their next start
21.09%

From this we see the biggest factor is the actual race time (with regards to 1200m races)

The top 10 race times over 1200m resulted in 5 winners next start.
50% strike rate.

And we haven't yet considered weight carried barrier position and most importantly track time or going.

We also haven't evaluated a second placed horse by various margins and given it a rating.

However I believe that times are irrelevant over distance races and sprints are pretty much the only category to benefit from this type of analysis.

[Chrome Prince]

Beton,
Don't get me wrong, I fully appreciate what you're trying to do, and you are quite correct, sometimes a different perspective can reveal things that seasoned investigators can't see because it's too obvious.
The best in the business is Mark Read followed by Sean Bartholomew, (at ratings that is).
They have cracked it big time.
I'm left struggling because the classes are a nightmare to line up.
I abandoned ratings and instead use professional handicappers and watch if there is money for their top rated horses.
This is the easiest approach for me.

[beton]

Hello Chrome Prince
Quote "I abandoned ratings and instead use professional handicappers and watch if there is money for their top rated horses.
This is the easiest approach for me." unquote. Even they get it wrong often.
What chance is there for us plebs? My thoughts are that with computers, too many zero in on the horse that on paper is the most likely winner. All using the same data and similar rating programs. True the majority of these do get up. But the price is slashed. On the TAB there is no way that backing these horses are viable. You have to cull the losers or get better prices. The only real way you are going to get better prices is decide your runner and to lock in the bookie early. Beton