• Hi Guest, The forum will be moving hosts on 26 July and as such will be closed from Midday until the move has completed.
    As we will be with new hosts it may take a while before DNS get updated so it could take while before you can get back on the forum.
    I think it will take at least 4 hours but could easily be 48!
    Ark Royal
  • There seems to be a problem with some alerts not being emailed to members. I have told the hosts and they are investigating.
  • Sorry for the ongoing issues that you may have been experiencing whilst using the forum lately

    It really is frustrating when the forum slows down or Server Error 500 pops up.

    Apparently the hosts acknowledge there is a problem.
    Thank you for using our services and sorry for the experienced delay!
    Unfortunately, these errors are due to a higher server load. Our senior department knows about the issue and they are working towards a permanent resolution of the issue, however, I'd advise you to consider using our new cPanel cloud solutions: https://www.tsohost.com/web-hosting


    I will have to investigate what the differences are with what We have know compared to the alternative service they want us to migrate to.
    Keep safe.
    AR

Chi, Exp and other statistical methods for evaluating systems

Chesham

Sire
Mind Mapping is a good technique when trying to come up,with a logical system or even analysing a race

 

newcomer

Yearling
Just to add to this, as I see I posted a very similar question without reading back through the comments (apologies).

I think chi and A/E help a lot, but don't fully account for selection bias. I've a number of systems in HRB that are very profitable over 10 years, but (except for luck) would probably make a loss in 2021. These are mostly ones that rely on picking individual humans/horses that have done well (eg, trainers/jockeys/stallions) rather than concepts (eg ratings/CD winners). Even with a good chi and A/E score, there's insufficient reason to think the same criteria and qualifications would do the same again the following year.

As an example, I ran with an idea that I think has some mileage. Entered the criteria, along with some basic control variables, and broke it down by the most successful trainers with my idea over 2011-2019. High ROI, high A/E, high strike rate. Results below for 2011-2019 very promising. Two bets a week, good SR, good ROI. Good A/E. Good chi. Etc etc.




Testing that system in 2020, you see the results below. Very poor by comparison, and certainly not reliable.


My point is that backtesting in some shape or form is the best I can think of (beyond proofing your system in the future). Run your system and its criteria to generate qualifiers as if it was a year ago, and see how it would have performed in 2020. HRB allows you to do that.
 

dave58

Administrator
Have to agree here.
Start with a basic idea, and only then think about where you go from there.
My way of thinking is that if I can't explain the logic behind my systems then they are likely to be worthless long term.
 

Emptymind

Yearling
Just realised why the stats are so high for day breaks- 1-7 for trainers. Winners without a penalty. If you take those out makes a big difference.
 

newcomer

Yearling
Just to clarify: I wasn't bringing together a random bunch of trainers. It was based on a logical idea (and some refinement has got a working system now with fewer qualifiers). But it highlights the issue.
 

Emptymind

Yearling
Looking at how these systems go forward at from past experience this doesn't reproduce. I've done as suggested here ie collect data from earlier years then test against 2020. I have an idea of what i want to do next. The Hypothesis is the data isn't reliable as it stands because of lets call it 'the flat track bully syndrome'. In this syndrome some sires have had easy races to perform in compared to others. I don't mean class of races. I mean they were racing in races against horses who's sires don't perform well with the above parameters.
So to correct this as much as i can. I'm going to select races where there is a minimum of 2 of the above sires who ran in such a race. I will then get a new ranking for the sires.
Note: I'm aware of the following. Sires who only have winners from 1 horse or 2 producing an unreliable win %. I can correct this.

I did something similar in the autumn of 2019. I did that with 2yr olds only. I was trying to rank the sires. For that data i found races with 2 or more sires, checked finishing position on a which sire beat which sire basis and re-ranked them. It was quite time consuming but not as bad as I imagined.

I'm just checking with folks on here for they're valued opinions insights or even a quicker/easier way to do this that i haven't thought of.

I also think a similar approach could be taken with jockey and trainer data.
For example a trainer could have high Win % that comes from good race placement, which doesn't reproduce for the trainer in more competitive affairs.
 
Last edited:

Emptymind

Yearling
Can you run by race class that would tell you more about a stables fire power and ambition
I think you can correct Trainer data by a certain degree by doing so. Im focusing on sires for now as that's my main interest on the aw also i tend only to bet in class 4-7 flat so it wont help me so much.
 

Emptymind

Yearling
Hi Emptymind Emptymind

I think you have fallen into the trap of Backfitting . The Aussie Rules, 0-9 already this year and the number of filters that you have, lack reasoning.
Track - Some sires perform better on certain all weather tracks that others. Not just surface but direction and time spent running round a bend etc.
Handicap- Non handicaps eliminated as stakes races have weights chosen by the trainers- producing erroneous data.
Number of runners - eliminating data with 4 or less runners again i consider this data erroneous- races not run at a true pace more often than the norm- objective opinion i know.
Age - eliminating 2 year old data - sires offspring perform differently at this age group than other years so I've eliminated those.
Distance- Sprint races chosen- 5 to 7f -reason i'm looking for sires who do well in sprint distances.
So they're the reasons why I've chosen the number of filters and reasons why.
As regards Backfitting sure- Iv'e generated a list of sires this way and chosen them. However i'm now going to see how those sires performed against each other on a race by race basis. Re-rank those sires as in the example above. Then see how that new ranking goes going forward.
 

Emptymind

Yearling
By the way i appreciate your help with this Chesham Chesham. I know more now. The information about pedigree has helped alot.
Ive been looking at pedigree and finding out more. Using Thoroughbred Horse Pedigree Query because of your comments previously.
I dont have enough information/resources to use Nicks in my analysis but i understand the concept. Indeed using pedigree analysis yesterday helped me find 28/1 winner Nicks No Wonder Yesterday. So Thank You
 

Emptymind

Yearling
Well I don't think your interested in this now but just in case out of those 9 wins for War Front. There was one win against a Lope De Vega from my list of sires. Which suggests the races were weak. Thanks again Chesham I will look at the info you've posted on here and try to find the sire data on the sheets.
 

dave58

Administrator
Just to clarify: I wasn't bringing together a random bunch of trainers. It was based on a logical idea (and some refinement has got a working system now with fewer qualifiers). But it highlights the issue.
Would you care to share the 'logical idea' with us
 
Top