True odds calculation
- wearthefoxhat
- Posts: 3221
- Joined: Sun Feb 18, 2018 9:55 am
That's the process for sure
I'd be tempted to Lay the slower starting T5 @ 3.25 in that one.
Sometimes keeping it simple is all that's needed.
You do not have the required permissions to view the files attached to this post.
"Using the sum and proportion approach..." It's a kinda primitive way to calc odds.If you'll try the method described in my link with a user defined parameters 10 and 90 you'll get 1,1 and 10. If you really hate any user involvement in calculation you can that try another approach: using excel use STANDARDIZE on rating and then use NORMSDIST than calc odds from numbers you got. It gives 1,2 and 6,3.firlandsfarm wrote: ↑Fri Dec 31, 2021 12:59 pmI'm afraid you are going to have to give up napshnap! That's the system that to me has no mathematical foundation. Why should adjusting your ratings by a random factor into an artificial structure and then pro-rata them give a reliable probability? Take a two runner race and give one runner a rating of 100 and the other a rating of 50. Using the sum and proportion approach would give odds of 1.50 and 3.00 respectively but I doubt you would find a bookmaker who would give you 1.50 on the favourite with such an ability advantage. I note that some contributors to that thread and the FlatStats thread referred to take a similar view as me.napshnap wrote: ↑Fri Dec 31, 2021 8:38 amSo, if this (viewtopic.php?p=275183#p275183) doesnt fit your desire then I give up).firlandsfarm wrote: ↑Fri Dec 31, 2021 3:52 am
No I don't think you misread that bit. there are many stabs at converting ratings to odds I just don't see them able to stand up to mathematical theory. By calculating I mean put a set of ratings into a spreadsheet and have it calculate the respective odds with no interaction with the user.
...
ps. AFAIR I overcomplicated it a lil bit and there is much simpler method in that thread.
The problem is being able to convert the odds compiler's judgement into an algorithm capable of being calculated. The human brain can think and reason, computers cannot until the analysts can tell the coders exactly how we think about and reason things in every combination of variables.
Closer to imaginable bookie prices innit?
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
If so that's skewed with small or large numbers, better to compare implied chance precentages. Tbh I never use odds in any calcs, its always percentages to keep the maths clean and simple and consistent over the whole ladder.
-
- Posts: 3217
- Joined: Thu Oct 24, 2019 8:25 am
- Location: Newport
There are a number of ways of doing it. In my example if Value < 1 then look to lay or > 1 then look to back.
For backing then use the highest number and for laying use the lowest.
Odds are always based on opinion, although in that race allot of people seem to get it wrong as the favourites sectional times were not great.
For backing then use the highest number and for laying use the lowest.
Odds are always based on opinion, although in that race allot of people seem to get it wrong as the favourites sectional times were not great.
- firlandsfarm
- Posts: 2720
- Joined: Sat May 03, 2014 8:20 am
Yep, it's what I call betting statistically. Finding angles (edges) that consistently give a profit over the long term but being prepared for losing runs in the short term. I do Monte Carlo scheduling of actual data to try and uncover the depth of the worst losing runs. Keep the faith.wearthefoxhat wrote: ↑Fri Dec 31, 2021 3:23 pmValue betting/laying can be frustrating, but equally rewarding when it clicks into place.
- firlandsfarm
- Posts: 2720
- Joined: Sat May 03, 2014 8:20 am
Glad to see you didn't give up!napshnap wrote: ↑Fri Dec 31, 2021 7:36 pm"Using the sum and proportion approach..." It's a kinda primitive way to calc odds.If you'll try the method described in my link with a user defined parameters 10 and 90 you'll get 1,1 and 10. If you really hate any user involvement in calculation you can that try another approach: using excel use STANDARDIZE on rating and then use NORMSDIST than calc odds from numbers you got. It gives 1,2 and 6,3.
Closer to imaginable bookie prices innit?
I don't hate user involvement, I'm just always looking for ways to automate and free myself up for other things.
I'll look into STANDARDIZE and NORMSDIST, never come across them before ... thanks for introducing them to me.
I think the interesting thing about your two reworks of my simple example is that you can experiment with different methods, back test the results and choose the one you are happiest with. The point of my (complicated) formula was it was a strict formula with no user involvement other than choosing the ratings system to use. Then like wearthefoxhat not to bet on the predicted winner but the ones that gives the best value be they back or lay.
As for the original subject of this thread I don't think you will ever get real true odds because there is always one factor that can never be evaluated ... is the selection "up for it"!
Hmm, I think we can calc odds from some metrics (ratings, weights) but it looks like there's no perfect universal method to do it. Even that method which I suggested (using Normal Distribution) is limited by Normal Distribution shape. But I think it's a closest one, cause if you take, for example, all horse racing or greyhounds racing ratings for a few months they will be following Normal Distribution.firlandsfarm wrote: ↑Sat Jan 01, 2022 11:41 amGlad to see you didn't give up!napshnap wrote: ↑Fri Dec 31, 2021 7:36 pm"Using the sum and proportion approach..." It's a kinda primitive way to calc odds.If you'll try the method described in my link with a user defined parameters 10 and 90 you'll get 1,1 and 10. If you really hate any user involvement in calculation you can that try another approach: using excel use STANDARDIZE on rating and then use NORMSDIST than calc odds from numbers you got. It gives 1,2 and 6,3.
Closer to imaginable bookie prices innit?
I don't hate user involvement, I'm just always looking for ways to automate and free myself up for other things.
I'll look into STANDARDIZE and NORMSDIST, never come across them before ... thanks for introducing them to me.
I think the interesting thing about your two reworks of my simple example is that you can experiment with different methods, back test the results and choose the one you are happiest with. The point of my (complicated) formula was it was a strict formula with no user involvement other than choosing the ratings system to use. Then like wearthefoxhat not to bet on the predicted winner but the ones that gives the best value be they back or lay.
As for the original subject of this thread I don't think you will ever get real true odds because there is always one factor that can never be evaluated ... is the selection "up for it"!
Probably not the right thread for this but..........
I have been looking into Linear Regression (thanks Gazuty for planting the seed to look at stats) and need a little advice.
From what I understand basically there are 3 numbers to look at.
The R Squared and the P Values for the Intercept and say my team Ranking. But some seem to contradict each other. My understanding for just one variable (team rank) an R value of over .6 is good and you want the P Values close to 0 preferably 0.005. So I did 2 tests and don't know which is best
First test I got
R 0.497
Intercept 1.89258E-08
Rank 0.0005
2nd test
R 0.619
Intercept 0.9676
Rank 3.87374E-05
Are both good or one or none? What confuses me is the 2nd test has a much better R but intercept is poor. I am thinking the best is the first one
Thanks for any help
I have been looking into Linear Regression (thanks Gazuty for planting the seed to look at stats) and need a little advice.
From what I understand basically there are 3 numbers to look at.
The R Squared and the P Values for the Intercept and say my team Ranking. But some seem to contradict each other. My understanding for just one variable (team rank) an R value of over .6 is good and you want the P Values close to 0 preferably 0.005. So I did 2 tests and don't know which is best
First test I got
R 0.497
Intercept 1.89258E-08
Rank 0.0005
2nd test
R 0.619
Intercept 0.9676
Rank 3.87374E-05
Are both good or one or none? What confuses me is the 2nd test has a much better R but intercept is poor. I am thinking the best is the first one
Thanks for any help
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
That's only half the story, you need to consider sample size and statistical significance. Also the consideration of the underlying strategy when it comes down to how much variability it should exhibit. You should also be separating your data into multiple random subsets so you can do proper in and out of sample testing to ensure you're not backfitting. Then using the difference between those tests as an indicator of how reliable it ought to be.
As soon as people find a few fancy sounding analytical methods they think that's it, but unless you fully understand the nuance and pitfalls they're more misleading than they are informative. Take a look at the Trading What I See thread. It's full of jargon, tech and impressive statistics lingo but it hasn't resulted in anything that's made any money because its implementation and interpretation have been lacking or rushed.
Historical price data and supplementary info are undoubtedly the key to finding a profit but like a lot of this it's deeper then it seems at first, so make sure you keep your brain engaged and think deeply about what it's not telling you or how it might be less than concrete instead of just taking the resulting R, P or whatever it is at face value.
As soon as people find a few fancy sounding analytical methods they think that's it, but unless you fully understand the nuance and pitfalls they're more misleading than they are informative. Take a look at the Trading What I See thread. It's full of jargon, tech and impressive statistics lingo but it hasn't resulted in anything that's made any money because its implementation and interpretation have been lacking or rushed.
Historical price data and supplementary info are undoubtedly the key to finding a profit but like a lot of this it's deeper then it seems at first, so make sure you keep your brain engaged and think deeply about what it's not telling you or how it might be less than concrete instead of just taking the resulting R, P or whatever it is at face value.
- ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
.... Which is best of your 2 tests?..... Impossible to tell from the info provided.
Thanks Shaun,ShaunWhite wrote: ↑Thu Jan 06, 2022 11:40 amThat's only half the story, you need to consider sample size and statistical significance. Also the consideration of the underlying strategy when it comes down to how much variability it should exhibit. You should also be separating your data into multiple random subsets so you can do proper in and out of sample testing to ensure you're not backfitting. Then using the difference between those tests as an indicator of how reliable it ought to be.
As soon as people find a few fancy sounding analytical methods they think that's it, but unless you fully understand the nuance and pitfalls they're more misleading than they are informative. Take a look at the Trading What I See thread. It's full of jargon, tech and impressive statistics lingo but it hasn't resulted in anything that's made any money because its implementation and interpretation have been lacking or rushed.
Historical price data and supplementary info are undoubtedly the key to finding a profit but like a lot of this it's deeper then it seems at first, so make sure you keep your brain engaged and think deeply about what it's not telling you or how it might be less than concrete instead of just taking the resulting R, P or whatever it is at face value.
Your right I have just started doing this and I have read Goats thread and tbh when I read most of that statistical stuff I had no clue what you guys were on about, if you could have seen into my mind then you would see tumble weed blowing across a desert. But now I am starting to understand what you were talking about so maybe time to re read that thread.
I am testing one thing at a time to see if it gets good results and will add more as time goes on, that test was a simple one to see if my Team ranking is an indication of goals for and against. To me it looks like it does show a link that the higher my ranking the more likely they are to score more, which I already knew but I wanted to prove it. But to me there is a difference one looked better than the other so wanted to continue with the best result but as you say not enough data there.
My plan is if my rankings are close then I can use that to then test what difference that will have on my Xgls, team selection, corners, free kicks. Maybe a ref? Or even are some pitches lower scoring than others by crowd size
My biggest issue IS sample size I only have 900 matches with all this info so am wary of that but I am recording stats for 5 European Leagues so that should build quickly. Then I can back test on the matches from today on, at the end of the season, if it is good may look at deploying it next season with small stakes.
But am finding this useful and enjoy doing it
I am just at a basic level and mostly just wanting to see if something is random. I compared my team rankings to team wealth and found that team wealth is slightly better than my rankings. So now I know this I can go back get past results get the wealth from years gone and back test. From that I now my ranking would have given similar results.
From that then I can look at things like does it matter to teams what time they play, days between matches, refs, how far they have travelled, weather conditions. These may all just be random and of no use but at least it will tell me if they are random, if not then that maybe an edge. But it is all pie in the sky at the moment, it's like what Peter said about the Oak tree, I am at my first choice of which way to go. Plus I am enjoying it
I know weather plays a part because back in the day when I played the wetter the better for me I loved the heavy going, hot days killed me and I wouldn't last the full 80. Heck I worked in a Cyclone (Bola) once, my mate said we not going out in that I said we have too, tuck a bit of shirt around it put your wet weather gear on and get it done
From that then I can look at things like does it matter to teams what time they play, days between matches, refs, how far they have travelled, weather conditions. These may all just be random and of no use but at least it will tell me if they are random, if not then that maybe an edge. But it is all pie in the sky at the moment, it's like what Peter said about the Oak tree, I am at my first choice of which way to go. Plus I am enjoying it
I know weather plays a part because back in the day when I played the wetter the better for me I loved the heavy going, hot days killed me and I wouldn't last the full 80. Heck I worked in a Cyclone (Bola) once, my mate said we not going out in that I said we have too, tuck a bit of shirt around it put your wet weather gear on and get it done