LLM Virtual Panel: 'Hippos' Preview all Premier Handicaps at Aintree

[matekus]

Our LLM Virtual Panel turns its attention to Aintree - three pundits, two dialectical interactions, and an occasional unhinged view.

You might enjoy the continuing chaos...
(https://vendire-ludorum.blogspot.com/20 ... ippos.html)

[matekus]

The 'Hippos' consensus selection got 'close, but no cigar!'
- 2nd: 8. Highlands Legacy 11/2

[Euler]

Have you tried to model the significance of each of the key factors in racing, like horse, jockey, progeny, course, distance and so on?

[matekus]

A little, here and there — enough to be dangerous, not enough to be smug!
Out of curiosity, are we talking straight feature selection for a logistic regression, or something a bit more adventurous?

[matekus]

The 'Hippos' Topham Handicap Chase preview...

[Euler]

A little, here and there — enough to be dangerous, not enough to be smug!
Out of curiosity, are we talking straight feature selection for a logistic regression, or something a bit more adventurous?

I'm thinking logistic regression. I'm really interested to see what the contributing factors are to a horse winning a race, but it seems difficult to nail down.

I really understand in play incredibly well now, but pre-off I'm sort of curious as to whether you can infer what the odds should be, given form, lines, progeny, trainer, jockey, course, distance, and so on. It's long been a target of mine to really nail that down.

[csewell1987]

A little, here and there — enough to be dangerous, not enough to be smug!
Out of curiosity, are we talking straight feature selection for a logistic regression, or something a bit more adventurous?

I'm thinking logistic regression. I'm really interested to see what the contributing factors are to a horse winning a race, but it seems difficult to nail down.

I really understand in play incredibly well now, but pre-off I'm sort of curious as to whether you can infer what the odds should be, given form, lines, progeny, trainer, jockey, course, distance, and so on. It's long been a target of mine to really nail that down.

I wouldnt necessarily say its difficult to nail down the attributes as such. You have the ones you mention above but also things like race class, rating, ground, etc. The difficulty comes with knowing how to weight the attributes in regards to the race that day. Different things matter depending on the setup of that race. You may have heavy ground which is going to have more of an effect on how horses will perform than say good to soft which is more optimum for most. So you would need to take that into account. Rating has most use in handicaps where it dictates how the horses are weighted. Something dropping down in mark might have reached something workable for today's contest or may just be falling off a cliff. Trip is another one where you have specialist distances like 7f, where previous form at this trip would be beneficial. I think this is why its so difficult to gauge what the odds should be of a given animal in a race.

[matekus]

Just to clarify - are we looking at it from a Technical Trading (TT) perspective or a Fundamental Handicapping (FH) one or both?

- TT: Predicting prices when the market is going in-play
- FH: Predicting the winner of the race

[Euler]

Just to clarify - are we looking at it from a Technical Trading (TT) perspective or a Fundamental Handicapping (FH) one or both?

- TT: Predicting prices when the market is going in-play
- FH: Predicting the winner of the race

I know to about a billion decimal places what the price should be in play, but it's more about standard handicapping. Trying to understand how the market is being priced and why.

I know the variables and how people arrive at decisions based upon individual variables, but I'm more interested in the weightings of them.

[matekus]

This DAG represents my starting position for a generic race.

[Morrid4]

Very interesting thread.....the most important one on here if someone really cracked it obviously!

I think the problem with weighting is how it can vary race to race. I wouldn't be so bold as to query any of the ratings shown in the graphic at all as they all seem sensible approaches.
However let's take one and look at jockey (5%). I think that is a fair average across most races all things being equal but then it does depend on the particular jockey. I have one in mind who in a negative sense would have far more effect on particular horses which I normally trade from a front runner trading perspective. Simply put, with any other jockey on that horse, then 5% would be the factor but with this jockey on, it suddenly it jumps up to become 50% of the overall and is probably a no trade for me.

I get that this is all average and over a lot of trades and attempting to put a system in which gets a long term edge but just an example of why I think it is so challenging to get that magic system which gives a rating and fires out the answer...we are human, it's what our quantitively oriented brains like? It's also possibly why qualitative approaches are tough, time consuming and probably pay off when applied consistently. Just my two cents.

[Euler]

This DAG represents my starting position for a generic race.

Thanks, that's very interesting.

[wearthefoxhat]

This DAG represents my starting position for a generic race.

Good post. That's as good as any out there.

Bill Benter/Alan Woods built their database(s) and drilled down to an Nth degree on each variable. TBH, I don't ever think their success could be replicated, however, through testing, it's possible a winning data model could be constructed.

A sort of rating/tissue/value comparison is then made on all runners, back all those that qualify, or, lay the most obvious ones out of line on the prices.

Nowadays, the books are a bit lazy pre-off, safe in the knowledge they can restrict a winning punter at any time, for any reason. Also, their line into the stables knowledge and betting behaviour is not like it used to be.

If approaching this from a database angle, don't forget to include the live market into the model. BB learned that early lesson, when he did...well the rest is history.

[csewell1987]

This DAG represents my starting position for a generic race.

Good post. That's as good as any out there.

Bill Benter/Alan Woods built their database(s) and drilled down to an Nth degree on each variable. TBH, I don't ever think their success could be replicated, however, through testing, it's possible a winning data model could be constructed.

A sort of rating/tissue/value comparison is then made on all runners, back all those that qualify, or, lay the most obvious ones out of line on the prices.

Nowadays, the books are a bit lazy pre-off, safe in the knowledge they can restrict a winning punter at any time, for any reason. Also, their line into the stables knowledge and betting behaviour is not like it used to be.

If approaching this from a database angle, don't forget to include the live market into the model. BB learned that early lesson, when he did...well the rest is history.

Thats where I struggle. Including the market without just becoming almost a mirror of it. Weighting it too little renders it almost redundant as the other attributes over power it. Weighting it too highly makes my implied probabilities line up closely with it. Guess its finding that middle ground which I havent been able to do yet programmatically. I take what my model spits out and do that manually currently adjusting them with what the market is saying.