How to calc true odds

[goat68]

What you could do is:

- Every dog in the race has run >= 5 times
- Get the CHESTER rating for each dog/race in the past and take an average.
- Repeat for each dog in the race
- Then do what Shaun suggested for working out true ish odds.

If the BF back price is > than your odds + margin then back it and if BF lay price < than your odds - margin then lay it.

Repeat for every qualifying race over a month and see what your P/L is.

That is scarily identical to what I was doing for my "dog model" last year, I even chose 5 races! it's obviously a far too obvious a method...!

[ShaunWhite]

"Shaun" didn't suggest it, ChatGPT did. I just pointed out that it was too simplistic. What I suggested was much more involved modeling 'types' of dog not specific dogs.

Dogs horses and football matches don't feature the same competitors often enough for historic details to be statistically significant. It's done by creating 'types' of competitors that individual ones are placed into. That's a technique that's been used since the 90s, pioneered by Bill Benter. It was a defining moment in the world of serious betting.

[Archery1969]

"Shaun" didn't suggest it, ChatGPT did. I just pointed out that it was too simplistic. What I suggested was much more involved modeling 'types' of dog not specific dogs.

Dogs horses and football matches don't feature the same competitors often enough for historic details to be statistically significant. It's done by creating 'types' of competitors that individual ones are placed into. That's a technique that's been used since the 90s, pioneered by Bill Benter. It was a defining moment in the world of serious betting.

Apologies, I stand corrected Sir.

[ShaunWhite]

Apologies, I stand corrected Sir.

no worries, happy to be confused with gpt.

[abgespaced]

No-code AI looming on the horizon doesn't make me feel good about any kind of coding to gamble haha

I've been working my ass off trying to get a fundamental greyhound ML model off the ground and see ChatGTP bust my balls in all of two minutes

[Anbell]

No-code AI looming on the horizon doesn't make me feel good about any kind of coding to gamble haha

I've been working my ass off trying to get a fundamental greyhound ML model off the ground and see ChatGTP bust my balls in all of two minutes

The next step will be to build a model that second-guesses chatGPT's models.

[abgespaced]

No-code AI looming on the horizon doesn't make me feel good about any kind of coding to gamble haha

I've been working my ass off trying to get a fundamental greyhound ML model off the ground and see ChatGTP bust my balls in all of two minutes

The next step will be to build a model that second-guesses chatGPT's models.

Inception!

[ilovepizza82]

The simple version..... (but simple obviously doesn't mean accurate otherwise there wouldn't be complex methods)

Q. In horse or greyhound racing, how do you calculated the odds for each selection based on a rating figure.

ChatGPT
A. Calculating odds from ratings is an intricate process, often involving a blend of mathematical models and subjective judgment. However, a simple and common approach to convert ratings into odds is the "probabilistic method". Here's a step-by-step guide for this method using horse or greyhound racing as an example:

Calculate the Sum of Ratings:
Sum the ratings of all the horses or greyhounds in the race.

For example, if you have 3 horses with ratings of 100, 80, and 60, the sum would be 240.

Calculate the Probability for Each Horse:
Divide each horse's individual rating by the sum of the ratings.

For the horse rated 100 in our example:
Probability = Horse's Rating / Sum of Ratings
= 100 / 240
= 0.4167 or 41.67%

Convert Probability to Decimal Odds:
Decimal odds = 1 / Probability

For the horse rated 100:
Decimal Odds = 1 / 0.4167
= 2.40

Convert Decimal Odds to Fractional Odds (if desired):
Subtract 1 from the decimal odds, and then express the result as a fraction.

For the horse rated 100:
Fractional Odds = (2.40 - 1)
= 1.40 or 7/5

So, the odds for the horse would be expressed as 7/5.

Repeat steps 2-4 for each horse or greyhound in the race.

It's important to note that this method provides a purely mathematical representation of odds based on ratings. In the real world, other factors like public perception, bookmaker margins, and external influences might alter the actual odds offered on selections. Furthermore, ratings themselves can be influenced by numerous factors like recent performance, track conditions, jockey/trainer statistics, and more.

I tried this method but i dont think it works.
When i calc the odds of all the runners in a race the odds look very much a like.
Like this:
3.82
3.77
2.81
3.57
3.3

They barely differ from each other.
I used 11:06 Perry Bar data, today.

[ShaunWhite]

I tried this method but i dont think it works.
When i calc the odds of all the runners in a race the odds look very much a like.
Like this:
3.82
3.77
2.81
3.57
3.3

They barely differ from each other.
I used 11:06 Perry Bar data, today.

Your calc must be wrong, those prices add up to more than 100%

But this is a very simplistic method, it's not going to get close to accurate price. That's why I posted that explanation of the method people actually use. Ie via modelling.

You should continue this conversion with ChatGPT.

[conduirez]

The simple version..... (but simple obviously doesn't mean accurate otherwise there wouldn't be complex methods)

Q. In horse or greyhound racing, how do you calculated the odds for each selection based on a rating figure.

ChatGPT
A. Calculating odds from ratings is an intricate process, often involving a blend of mathematical models and subjective judgment. However, a simple and common approach to convert ratings into odds is the "probabilistic method". Here's a step-by-step guide for this method using horse or greyhound racing as an example:

Calculate the Sum of Ratings:
Sum the ratings of all the horses or greyhounds in the race.

For example, if you have 3 horses with ratings of 100, 80, and 60, the sum would be 240.

Calculate the Probability for Each Horse:
Divide each horse's individual rating by the sum of the ratings.

For the horse rated 100 in our example:
Probability = Horse's Rating / Sum of Ratings
= 100 / 240
= 0.4167 or 41.67%

Convert Probability to Decimal Odds:
Decimal odds = 1 / Probability

For the horse rated 100:
Decimal Odds = 1 / 0.4167
= 2.40

Convert Decimal Odds to Fractional Odds (if desired):
Subtract 1 from the decimal odds, and then express the result as a fraction.

For the horse rated 100:
Fractional Odds = (2.40 - 1)
= 1.40 or 7/5

So, the odds for the horse would be expressed as 7/5.

Repeat steps 2-4 for each horse or greyhound in the race.

It's important to note that this method provides a purely mathematical representation of odds based on ratings. In the real world, other factors like public perception, bookmaker margins, and external influences might alter the actual odds offered on selections. Furthermore, ratings themselves can be influenced by numerous factors like recent performance, track conditions, jockey/trainer statistics, and more.

I tried this method but i dont think it works.
When i calc the odds of all the runners in a race the odds look very much a like.
Like this:
3.82
3.77
2.81
3.57
3.3

They barely differ from each other.
I used 11:06 Perry Bar data, today.

You will have to work out how to do the odds, but remember to achieve a 100% overround on greyhounds in a six dog race if all had equal chance then each dog would all have to be 5/1 you are never ever going to see a bookmaker offering that. The equivalent on Belfair would be every dog priced up at 6.0.

When you do work out how to do the odds the following may help you with your pricing model.

There are many variables in greyhound racing you need to factor in that you will not know, these include, the state of the floor of the traps, the speed that the hare is set for at the start of a race, higher grades have faster dogs. The skill of the hare controller who sits high up in the stands and can speed the hare up or slow it down to keep it so far in front of the dogs, if the hare is too close to the dogs, the leading greyhounds move towards it and can run wide around a bend. Another factor is the weight of the greyhound and what is its optimal racing weight. Another is bitches coming out of season normally they are back up to their best after 16 weeks, but that's just an average.

Its not all doom and gloom though, do not base any model on which greyhound you just think will win the race, there is are too many variables, you must break the race down. Most races are won or lost at the first bend, plan your model around this, which dogs normally make the bend in front, are there going to be dogs hitting the first bend at the same time as others try and work out which of these dogs are going to come off worse, see if there is the chance that a dog that finishes well can slip round the inside at the first bend if there's trouble, it's possible they could make up three or four places.

If the races are over more than two bends, the next but not so important pinch point is the third bend, so if your model predicts the order of greyhounds after the first two bends say 235641, which of them is likely to hit the third bend fin front, in the vast majority of races it will be one in the first three, that will be 2, 3 or 5 in the prediction above.

Finally which greyhound is most likely to win from your model from the last bend to the line.

Put this data together and you should be able to make reasonable predictions about what will be a good price

Your model in a four bend race should try and predict the order for each stage of the race following each of the points above. It will be something like 235641 at the first bend, 325614 at the second bend and 356214 at the line.

The minimum data you will need is trap to line time, the position the greyhound usually achieves at the first bend, does it usually make up ground between the second and third bend, does it usually make up or lose ground from the fourth bend to the line. All these figures are on a standard greyhound form guide. But you could easily make your model more complex with past form between the greyhounds, bitches season dates, has the dog been backed etc.

[ilovepizza82]

The simple version..... (but simple obviously doesn't mean accurate otherwise there wouldn't be complex methods)

Q. In horse or greyhound racing, how do you calculated the odds for each selection based on a rating figure.

ChatGPT
A. Calculating odds from ratings is an intricate process, often involving a blend of mathematical models and subjective judgment. However, a simple and common approach to convert ratings into odds is the "probabilistic method". Here's a step-by-step guide for this method using horse or greyhound racing as an example:

Calculate the Sum of Ratings:
Sum the ratings of all the horses or greyhounds in the race.

For example, if you have 3 horses with ratings of 100, 80, and 60, the sum would be 240.

Calculate the Probability for Each Horse:
Divide each horse's individual rating by the sum of the ratings.

For the horse rated 100 in our example:
Probability = Horse's Rating / Sum of Ratings
= 100 / 240
= 0.4167 or 41.67%

Convert Probability to Decimal Odds:
Decimal odds = 1 / Probability

For the horse rated 100:
Decimal Odds = 1 / 0.4167
= 2.40

Convert Decimal Odds to Fractional Odds (if desired):
Subtract 1 from the decimal odds, and then express the result as a fraction.

For the horse rated 100:
Fractional Odds = (2.40 - 1)
= 1.40 or 7/5

So, the odds for the horse would be expressed as 7/5.

Repeat steps 2-4 for each horse or greyhound in the race.

It's important to note that this method provides a purely mathematical representation of odds based on ratings. In the real world, other factors like public perception, bookmaker margins, and external influences might alter the actual odds offered on selections. Furthermore, ratings themselves can be influenced by numerous factors like recent performance, track conditions, jockey/trainer statistics, and more.

I tried this method but i dont think it works.
When i calc the odds of all the runners in a race the odds look very much a like.
Like this:
3.82
3.77
2.81
3.57
3.3

They barely differ from each other.
I used 11:06 Perry Bar data, today.

You will have to work out how to do the odds, but remember to achieve a 100% overround on greyhounds in a six dog race if all had equal chance then each dog would all have to be 5/1 you are never ever going to see a bookmaker offering that. The equivalent on Belfair would be every dog priced up at 6.0.

When you do work out how to do the odds the following may help you with your pricing model.

There are many variables in greyhound racing you need to factor in that you will not know, these include, the state of the floor of the traps, the speed that the hare is set for at the start of a race, higher grades have faster dogs. The skill of the hare controller who sits high up in the stands and can speed the hare up or slow it down to keep it so far in front of the dogs, if the hare is too close to the dogs, the leading greyhounds move towards it and can run wide around a bend. Another factor is the weight of the greyhound and what is its optimal racing weight. Another is bitches coming out of season normally they are back up to their best after 16 weeks, but that's just an average.

Its not all doom and gloom though, do not base any model on which greyhound you just think will win the race, there is are too many variables, you must break the race down. Most races are won or lost at the first bend, plan your model around this, which dogs normally make the bend in front, are there going to be dogs hitting the first bend at the same time as others try and work out which of these dogs are going to come off worse, see if there is the chance that a dog that finishes well can slip round the inside at the first bend if there's trouble, it's possible they could make up three or four places.

If the races are over more than two bends, the next but not so important pinch point is the third bend, so if your model predicts the order of greyhounds after the first two bends say 235641, which of them is likely to hit the third bend fin front, in the vast majority of races it will be one in the first three, that will be 2, 3 or 5 in the prediction above.

Finally which greyhound is most likely to win from your model from the last bend to the line.

Put this data together and you should be able to make reasonable predictions about what will be a good price

Your model in a four bend race should try and predict the order for each stage of the race following each of the points above. It will be something like 235641 at the first bend, 325614 at the second bend and 356214 at the line.

The minimum data you will need is trap to line time, the position the greyhound usually achieves at the first bend, does it usually make up ground between the second and third bend, does it usually make up or lose ground from the fourth bend to the line. All these figures are on a standard greyhound form guide. But you could easily make your model more complex with past form between the greyhounds, bitches season dates, has the dog been backed etc.

Thanks ! I will try that.
Appreciate your help.

[Tiesto13]

So quite early on in my trading career I seem to have ended up at the greyhounds…

When trying to learn how to trade pre off horses, I’ve read a lot of posts by people arguing about whether you can only profit trading if you know what a value bet is and isn’t, and those who argue that you can profit by taking advantage of short term market movements whilst having no idea whether the two opposing sides of the trade are value or not. Anyway, without wanting to start that discussion here, it did make me think that being able to approximate what the BSP for a horse should be would be the best way to make money trading. I realise that this is incredibly complicated and probably the life goals of people who are a) far more intelligent than me and b) know far more about horse racing than me.

Anyhow, I have managed to do something similar in cricket where whilst I’m not sure I can calculate the ‘true odds’, I can certainly calculate better odds than the bookies can for a number of markets. These models however are pretty simple with a small number of variables.

I did start to think how might I approach it for horse racing, and whilst one way is to look at all the race card data/form etc, I then thought about why not use the BSP data, as that is effectively the result of the market working out how good every horse is against every other horse in a race already and putting it into nice probabilities for me and on one handy download site where I can get it updated every day.

The issue I then realised was that handicapping, going, jockey etc are all variables that will vary wildly race to race and I’d effectively have to go back and weight every BSP for each of these items before I can even begin to think about using it. Which is difficult when I have no idea how much weight to give to any of these items.

So dead end. Or so I thought. But then I realised that the BSP data also has greyhound data in it. And I did a bit of research today, and greyhounds seems to have far less variables race to race. So I have to do far less adjustment to that initial BSP before using it to rank the dogs in each race.

So, my idea is to effectively use BSP to rank every dog in a race. The next step theoretically would be to give every race a value depending on the quality of dog in it. For example if dog 1 was a favourite in a low grade race because the competition was low, it should necessarily rank higher than an outsider in a higher grade. If I could come up with a way of modelling the quality of the race, then finding an overall ranking for each dog in a new race would be theoretically possible…

This is a slightly different way of looking at it from one mentioned earlier in this thread (by ShawnWhite) which was to effectively come up with different types of competitors based on the historic pricing data. Would like to take a look at that, is there any decent reading on this topic?

The other method obviously is to start collecting greyhound data, creating a model from first principles (speed etc) and then back testing it against the BSP data I have to see if it works or not and then tweak as necessary.

Also, as an aside, is it me or is the greyhound BSP data quite messy with several csvs that have issues?

[Michael5482]

So quite early on in my trading career I seem to have ended up at the greyhounds…

When trying to learn how to trade pre off horses, I’ve read a lot of posts by people arguing about whether you can only profit trading if you know what a value bet is and isn’t, and those who argue that you can profit by taking advantage of short term market movements whilst having no idea whether the two opposing sides of the trade are value or not. Anyway, without wanting to start that discussion here, it did make me think that being able to approximate what the BSP for a horse should be would be the best way to make money trading. I realise that this is incredibly complicated and probably the life goals of people who are a) far more intelligent than me and b) know far more about horse racing than me.

Anyhow, I have managed to do something similar in cricket where whilst I’m not sure I can calculate the ‘true odds’, I can certainly calculate better odds than the bookies can for a number of markets. These models however are pretty simple with a small number of variables.

I did start to think how might I approach it for horse racing, and whilst one way is to look at all the race card data/form etc, I then thought about why not use the BSP data, as that is effectively the result of the market working out how good every horse is against every other horse in a race already and putting it into nice probabilities for me and on one handy download site where I can get it updated every day.

The issue I then realised was that handicapping, going, jockey etc are all variables that will vary wildly race to race and I’d effectively have to go back and weight every BSP for each of these items before I can even begin to think about using it. Which is difficult when I have no idea how much weight to give to any of these items.

So dead end. Or so I thought. But then I realised that the BSP data also has greyhound data in it. And I did a bit of research today, and greyhounds seems to have far less variables race to race. So I have to do far less adjustment to that initial BSP before using it to rank the dogs in each race.

So, my idea is to effectively use BSP to rank every dog in a race. The next step theoretically would be to give every race a value depending on the quality of dog in it. For example if dog 1 was a favourite in a low grade race because the competition was low, it should necessarily rank higher than an outsider in a higher grade. If I could come up with a way of modelling the quality of the race, then finding an overall ranking for each dog in a new race would be theoretically possible…

This is a slightly different way of looking at it from one mentioned earlier in this thread (by ShawnWhite) which was to effectively come up with different types of competitors based on the historic pricing data. Would like to take a look at that, is there any decent reading on this topic?

The other method obviously is to start collecting greyhound data, creating a model from first principles (speed etc) and then back testing it against the BSP data I have to see if it works or not and then tweak as necessary.

Also, as an aside, is it me or is the greyhound BSP data quite messy with several csvs that have issues?

Please take Kai's advice on the previous thread and concentrate on one sport to trade, one strategy and master it. It's hard enough flipping a profit in one sport let alone four at once as a beginner.

Most of us have been there me included starting with football got nowhere thought racing will be easier moved to tennis done nothing then messing about with them all before going back to football for example. Your spreading yourself to thin and just wasting time, effort and possibly money.

Doing what your doing is like trying to cook a 2 star Michelin Star menu for 50 when you can't make an omelette. I hope you get there, I like to see people to do well but your going the hard way about it.

[firlandsfarm]

The simple version..... (but simple obviously doesn't mean accurate otherwise there wouldn't be complex methods)

Q. In horse or greyhound racing, how do you calculated the odds for each selection based on a rating figure.

ChatGPT
A. Calculating odds from ratings is an intricate process, often involving a blend of mathematical models and subjective judgment. However, a simple and common approach to convert ratings into odds is the "probabilistic method". Here's a step-by-step guide for this method using horse or greyhound racing as an example:

Calculate the Sum of Ratings:
Sum the ratings of all the horses or greyhounds in the race.

For example, if you have 3 horses with ratings of 100, 80, and 60, the sum would be 240.

Calculate the Probability for Each Horse:
Divide each horse's individual rating by the sum of the ratings.

For the horse rated 100 in our example:
Probability = Horse's Rating / Sum of Ratings
= 100 / 240
= 0.4167 or 41.67%

Convert Probability to Decimal Odds:
Decimal odds = 1 / Probability

For the horse rated 100:
Decimal Odds = 1 / 0.4167
= 2.40

Convert Decimal Odds to Fractional Odds (if desired):
Subtract 1 from the decimal odds, and then express the result as a fraction.

For the horse rated 100:
Fractional Odds = (2.40 - 1)
= 1.40 or 7/5

So, the odds for the horse would be expressed as 7/5.

Repeat steps 2-4 for each horse or greyhound in the race.

It's important to note that this method provides a purely mathematical representation of odds based on ratings. In the real world, other factors like public perception, bookmaker margins, and external influences might alter the actual odds offered on selections. Furthermore, ratings themselves can be influenced by numerous factors like recent performance, track conditions, jockey/trainer statistics, and more.

Wow. This loooks quite easy to do. Ill try that one as well. Thanks !

As Shaun says in response to you, easy does not mean correct! I have spent years trying to find a way to match ratings to odds and have never found what I would consider to be an accurate approach. Remember also the conversion to odds, no matter what the theory will only be as good as the ratings themselves. But try the ChatGPT method with a few historic races, rate the runners however you will and calculate the 'ChatGPT odds' for each runner. What you should instantly notice is that a small differential in the rating between runners will give rise to a large differential in odds.

Take the 2:20 at Cheltenham yesterday ...

Homme Public OR 138 BSP 7.67 RatOdds 4.01
Petit Tonnerre OR 139 BSP 6.20 RatOdds 3.99
Mighty Tom OR 143 BSP 3.43 RatOdds 3.87
JPR One OR 134 BSP 2.28 RatOdds 4.13
Total 554

I'm not looking to compare which to back and which to lay nor start a discussion on the value of the Official Rating system but look at the tightness of the calculated 'ratings odds' using the ChatGPT calculation compared to the actual odds. A small difference in the rating can result in a disproportional difference in the actual odds. My advice if you want to follow the ChatGPT approach do paper trade your parameters and build a good database for analysis.