Market Pricing - Discussion - Racing Talk - Racehorse TALK
harm-plan
harm-plan

Racehorse TALK



Market Pricing - Discussion - Racing Talk - Racehorse TALK

Author Topic: Market Pricing - Discussion  (Read 3843 times)

0 Members and 1 Guest are viewing this topic.

Online wily ole dog

  • Group 1
  • User 218
  • Posts: 30455
« 2022-Feb-16, 07:05 PM Reply #25 »
Wily, I sure did reduce my bet.
The original price and continued money for my second selection’s stablemate worried me most. I was happy to see Demando stay strong at the same time though. The favourite was a little easy, so that was ok. Late money on the top weight was ok too.

So at the end I just had ‘enough’ on the winner and saved on the second horse each way. I was a little weak, understandably considering my fancy was first up, but did well enough that I have no real regrets. Rapt that my market was right, considering I went public with it.

Cheers

Yes indeed, well done  :thumbsup:

Online jfc

  • Group 1
  • User 723
  • Posts: 7726
« 2022-Feb-17, 07:14 PM Reply #26 »
jfc, it really is fairly simple and gets easier as you use it. I learned this approach from an old time Melbourne bookie. As Gintara says, it really does convert your assessments of each horse into an accurate market (in terms of reflecting your thoughts):
Thanks Rad,

I joined in 2009 and you were here before me. but I don't think our paths ever crossed before.

There are more than enough tells to see that you are near the top of the field at what you do, just as there are enough tells to see that fours doesn't rate.

But I've always approached this from a mathematical angle and your pricing method, along with that of Don Scott, Dom Bierne and many others of that era are based on unsound mathematical reasoning.

I've spent some time today trying to find an effective way to illustrate this for the many here who still haven't realised that, and just came up with this.

If the chances of A beating B = 5/9,
and B beating C = 5/9:

What are the chances of A beating C?
What is the true Market for the Trio?

While these may seem like simple problems, and a bit of number crunching may provide some answers, this is far from the case.

Just look up Efron's dice to be stunned.

There it shows how C can actually beat A!

When A, B and C are intransitive dice.

Efron's case is even weirder!

The moral here is being told the chances of A beating B is next to useless by itself.

Instead you need to know the performance distributions of A, B, C and the rest follows.

So just having some Ratings Numbers is not enough, you need to know whether this is (say) a Normal Distribution, and if so what is the Variance.

Now we have already been through this last June, but that thread may not be that readable thanks to an intrepid bunch of Buttinskis bent on gumming up the works with their cloaca.

But despite this we ended up with the situation where Carey and Bascoe (both closer to the Rad end of the spectrum than mine, and who have an abundant scope for improving their Maths/Programming skills) managed to develop Monte Carlo simulators suitable for converting Ratings to Odds.

It's not hard.

=NORMINV(Ran,Rating,STD)
gives you a Random Sample for a Runner's performance

Sort those Samples and Tally up all the positions.

Repeat 100,000 times or whatever to get Odds for each Runner and position.

Anybody interested in Market Pricing should revisit that thread and try to learn something from it.










https://www.racehorsetalk.com.au/racing-talk/sky-ratings-as-per-jfc/

Offline fours

  • Group 1
  • User 704
  • Posts: 9960
« 2022-Feb-17, 07:29 PM Reply #27 »
  :lol:   :lol:   :lol:

Thanks for the compliment jfc.

Fours

Offline Rad

  • Group3
  • User 168
  • Posts: 924
« 2022-Feb-17, 08:04 PM Reply #28 »
Thanks for the post, jfc. I’ve done some maths and statistics in my day, so I’m on board with you.

A few things:

My approach was developed by a bookmaker. His aims were those of a bookmaker. Pick up a percentage of the action, irrespective of the outcome of the race .. or …. Give me a price where the odds are in my favour and, in the long run, I’ll finish in front.

The biggest advantage that I’ve got as a punter is that I can choose the battlefield… or I can choose to run and play another day. With this in mind, I can use the bookie’s method in my favour. I can look at Saturday’s races and choose to wait for another day - he can’t. I can play in race 1 and wait until the last to bet again. Bet when you are confident ‘enough’ to believe you can win, knowing you’ll be wrong plenty of times.

I really like the concept of normal distribution. In coming up with a price for a horse, I know that the likely performance of each horse can vary, sometimes wildly. You also realise that the distribution is almost certainly not normal around the ‘calculated’ / expected performance. Most horses have far greater potential downside, although a young improver can have very significant upsize. How will a first upper perform? Etc.

As much as anything, this determines a betting race vs a race to leave alone. How confident can you be in this race?

What I really do is come up with an order of ‘winning chance’ - but this is overlaid by a realisation of the likely variability of the ranking. It works for me. Not always right, hopefully right often enough.

Offline fours

  • Group 1
  • User 704
  • Posts: 9960
« 2022-Feb-17, 08:42 PM Reply #29 »
 :chin:

I seem to recall some one on this forum has mentioned variance before....

Fours

Offline Wenona

  • VIP Club
  • Group 1
  • User 175
  • Posts: 7828
« 2022-Feb-18, 05:44 PM Reply #30 »
If we are expressing future performance of three runners with a Rating, Standard Deviation and Normal Distribution they will never be Intransitive will they?

If A has a greater than 50% chance of beating B and B has a greater than 50% chance of beating C you can end up with a scenario where C is more likely to beat A than B is but the probability that C will beat A will never be greater than 50%, will it?


Offline Wenona

  • VIP Club
  • Group 1
  • User 175
  • Posts: 7828
« 2022-Feb-18, 05:57 PM Reply #31 »
I agree with rad, when determining a market from a ratings method the issue I have always struggled with is the fact the distribution will be either positively or negatively skewed around the rating and the std dev is thus greater than for a runner with more consistent form and less unknowns. As a work around I think we all trim the rating if we think it is positively skewed (ie long tail on the down side?) eg if an unknown factor (unproven at distance, drop off in form last or last couple of runs) and give a bonus if we feel the raw rating is negatively skewed (ie long tail on the up side?) eg unknown scope for improvement (age improvement, fitness improvement, experience improvement, favourable change of conditions). 

I know there's methods that can employed if the mean and std dev are based on a 30+ sample size but I don't think that applies here - be happy to be taught another lesson from jfc if he had any suggestions.
« Last Edit: 2022-Feb-18, 06:01 PM by Wenona »

Online jfc

  • Group 1
  • User 723
  • Posts: 7726
« 2022-Feb-18, 07:02 PM Reply #32 »
If we are expressing future performance of three runners with a Rating, Standard Deviation and Normal Distribution they will never be Intransitive will they?
Correct. But here you know everything about the RVs.

But I chose those weird dice to be a dramatic illustration that if all you are told are H2H probabilities you cannot even infer ranking, let alone quantify H2H Odds, let alone frame a full Market.

But that's what Don Scott does. All he has are some H2H Odds. Nowhere does he indicate what the distribution is - let alone its Variance.

But even if you assume he was talking about a Normal distribution, and calculated the implied Variance, all the adding crap he then cranks out is deranged onanism.

By contrast the Monte Carlo I described is easy and correct within the given parameters.



Online Peter Mair

  • Group 1
  • User 326
  • Posts: 6586
« 2022-Feb-18, 08:23 PM Reply #33 »


........... THIS IS GETTING TOO COMPLEX

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-02, 12:09 PM Reply #34 »
Wow - a useful topic to read and work on first time in a while.  I agree the Monte Carlo sims are a great way to go and move far beyond a Bayesian approach.  I did a lot of work on it last year and very happy with the results.

bascoe

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 09:00 AM Reply #35 »
jfc a question if I may..

getting back to the thread from last year, that you referenced in this thread, we had runners rated on an integer scale along the lines 100,98,97,96,95,80,100

Running the monte carlo sim using the code I developed this worked well and provides a series of probabilities that aligns within the expected range.

My question is how do I run the simulation using a different rating scale.  For example in a horse racing scenario, the above horses maybe rated according to a handicapping weights:
61.5
59.5
59
58
55.5
49
61.5

Similarly, in a greyhound scenario, dog maybe rated by predicted run time:
17.05
17.12
17.35
17.55
and so on

I have run the monte carlo using such figures as the 'rating' but the probabilities are not valid. 

There must be a way to define the rating range so the inverse normal 'score' uses the different range in place

bascoe


Online jfc

  • Group 1
  • User 723
  • Posts: 7726
« 2022-Mar-03, 10:12 AM Reply #36 »
If A has an 80% chance of beating B, then

=NORMINV(0.8, 0, SQRT(2))

tells you their Ratings difference

=1.190232


if their STDs = 1

So if your data shows an X seconds difference wins 80%

your STD = X/1.190232

Presto!

I just plucked 80% out of the air, tailor the above whatever H2H probability and Gap you data has.

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 12:20 PM Reply #37 »
Thanks jfc

Seems I need to get a grip on a few missing steps..

I am using the MonteCarlo to arrive at a set of probabilities for each runner.  Once assessed, all I have is a list of 8 times:

For example, at Sandown dogs, Race 12 tonight I have the following predicted run times:

1 30.45
2 30.10
3 30.10
4 30.05
5 29.95
6 30.00
7 29.90
8 29.80

from this data I do not yet have the percentage chance of any runner going head to head;
What I was hoping to do was creating a SCORE for each runner using these times as per the last example when I used RATING:

TabNo     Rating   STD   Rand()           Score (InvNorm)         Position
1          100             7   0.439728975  98.93840523   1
4          50             7   0.028932513   36.7229665   2
7          40             7   0.158895514   33.00694708   3
6          10             7   0.784290475   15.50735972   4
5          5             7   0.658050135       7.85003181   5

In this instance I will be substituting RATING with TIME...

How do I attack the problem from this perspective?
« Last Edit: 2022-Mar-03, 12:29 PM by bascoe »

Online jfc

  • Group 1
  • User 723
  • Posts: 7726
« 2022-Mar-03, 03:10 PM Reply #38 »
I can't find the Variance of dog times anywhere.

And you don't have the dog equivalent of the Don Scott table (below) that would allow you to determine the Variance.

So I don't know to help further.

If you've kept a history or predicted times along with actual finish positions you should be able to generate your own table.

https://www.championbets.com.au/betting-education/framing-market-basis-value-betting/

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 04:31 PM Reply #39 »
I do have my own table :

Expected_Margin   Decimal_Odds
0   1.00
0.01   0.90
0.02   0.90
0.03   0.80
0.04   0.70
0.05   0.70
0.06   0.69
0.07   0.68

and so on so I can post the variances along with the times.  Wasn't sure what I needed to include. 

So for this race...

Tab Time Margin DecOdds
1 30.45 0.64 0.04
2 30.10 .03 0.22
3 30.10 .03 0.22
4 30.05 .025 0.26
5 29.95 .013 0.53
6 30.00 .020 0.35
7 29.90 0.1 0.65
8 29.80 0.0 1.0

Thanks again

« Last Edit: 2022-Mar-03, 04:44 PM by bascoe »

Offline Wenona

  • VIP Club
  • Group 1
  • User 175
  • Posts: 7828
« 2022-Mar-03, 04:45 PM Reply #40 »
Hi

For what it's worth - I use a model with a std dev of 0.13 seconds. I only do races between 320m & 460m and only races of Grade 5 or better and it serves me pretty well.
I also model expected first sectional.

I only use the model price as one tool in my analysis and don't slavishly back all runners over the required price. Obviously race pattern is very important - if I price a dog as favourite and it's best couple of times are when its led and tonight I have it beaten for early speed - I probably won't back unless the price is ridiculous.
I've been using dog models for many years but I still find thinking a superior go to.

Regarding the example race at 100% my model would price the race as follows:

#1 - virtually none
#2 - $90.92
#3 - $90.92
#4 - $35.08
#5 - $7.69
#6 - $15.42
#7 - $4.33
#8 - $1.91

I ran a MC - 100,000 trials and got:

#1 - none
#2 - $78.06
#3 - $75.59
#4 - $33.26
#5 - $8.41
#6 - $16.42
#7 - $4.78
#8 - $1.80







Offline Rad

  • Group3
  • User 168
  • Posts: 924
« 2022-Mar-03, 05:31 PM Reply #41 »
Sorry to post in the middle of this thread progressing in an interesting way ...

Since my last post, I've followed up a lot of background information as suggested by jfc and others.

It is wrong to compare what I do to develop my market pricing with the Don Scott method. I read his book, took a lot of interesting things from the book,referred back to his book several times over the decades .... but I never ever accepted his market pricing methods or calculations and never ever used them. I could never see the logic of the pricing, but I did like the approach which gathered data and tried to make sense of the data.

I don't see any issues with the Efron's Dice information. In developing my view as to the relative winning chance of each horse, I have to work through conflicts and inconsistencies in most, if not all, races. Market framers would understand the importance of looking for hints and information to split the ratings of each and every horse in a race, even looking for a reason to push a horse from 100/1 out to 200/1.

It is wrong to judge my approach and opinioning that it is comparing A to B, B to C ... but not A to C. You have to work in depth with each and every runner if you are to develop a valuable market and a good assessment of each runner. The stuff I do is not simplistic.

The area that is perhaps a bit flaky is the process of converting the assessment of each runner into a market price. In fact, I have adjusted it a little over the years and I'm happier with it now. But the hard bit is the assessment and that has stood the test of time. The reality is, in the true test of any approach > $$, the pricing approach works too

Online Peter Mair

  • Group 1
  • User 326
  • Posts: 6586
« 2022-Mar-03, 07:11 PM Reply #42 »

.
« Last Edit: 2022-Mar-03, 08:07 PM by sobig »

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 07:41 PM Reply #43 »
PM please go away

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 07:51 PM Reply #44 »
For what it is worth - and I have many adjustments to my raw times based on RMW - early speed and so on my own prices for this race a
1 83.5
2 14.3
3 14.3
4 12.2
5 7.8
6 9.1
7 4.8
8 3.1

Just  waiting for jfc now to revert on my original proposition.
« Last Edit: 2022-Mar-03, 08:32 PM by bascoe »

Online Peter Mair

  • Group 1
  • User 326
  • Posts: 6586
« 2022-Mar-03, 08:10 PM Reply #45 »
Peter
You have nothing to contribute here so please refrain from posting on this thread
« Last Edit: 2022-Mar-03, 09:34 PM by sobig »

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-03, 08:54 PM Reply #46 »
.
« Last Edit: 2022-Mar-04, 06:46 AM by bascoe »

Offline bascoe

  • Open
  • User 2568
  • Posts: 188
« 2022-Mar-04, 06:50 AM Reply #47 »
Wenona,

You wrote that "I ran a MC - 100,000 trials"

Did you use run times or convert them to ratings?

If run times were used, was your RND() in the range of 0-1?

The answer to these questions is what I am chasing

bascoe

Online wily ole dog

  • Group 1
  • User 218
  • Posts: 30455
« 2022-Mar-04, 08:14 AM Reply #48 »
Sorry to post in the middle of this thread progressing in an interesting way ...

Since my last post, I've followed up a lot of background information as suggested by jfc and others.

It is wrong to compare what I do to develop my market pricing with the Don Scott method. I read his book, took a lot of interesting things from the book,referred back to his book several times over the decades .... but I never ever accepted his market pricing methods or calculations and never ever used them. I could never see the logic of the pricing, but I did like the approach which gathered data and tried to make sense of the data.

I don't see any issues with the Efron's Dice information. In developing my view as to the relative winning chance of each horse, I have to work through conflicts and inconsistencies in most, if not all, races. Market framers would understand the importance of looking for hints and information to split the ratings of each and every horse in a race, even looking for a reason to push a horse from 100/1 out to 200/1.

It is wrong to judge my approach and opinioning that it is comparing A to B, B to C ... but not A to C. You have to work in depth with each and every runner if you are to develop a valuable market and a good assessment of each runner. The stuff I do is not simplistic.

The area that is perhaps a bit flaky is the process of converting the assessment of each runner into a market price. In fact, I have adjusted it a little over the years and I'm happier with it now. But the hard bit is the assessment and that has stood the test of time. The reality is, in the true test of any approach > $$, the pricing approach works too


Many ratings don’t do this from my limited understanding of ratings numbers. Especially “legitimate” finishing position in a given race

Offline Wenona

  • VIP Club
  • Group 1
  • User 175
  • Posts: 7828
« 2022-Mar-04, 10:23 AM Reply #49 »
bascoe, I just ran 100,000 trials with the times and a std dev of 0.13 - the percentages are the frequency the dog finished on top.


BACK TO ALL TOPICS
Sitemap