Arbusers

can't find the formula to calculate over 2,75 from given odds of over 2,5 and over 3,0 (example below Book around 99,8%)
my formula, 50% o2,5 + 50% o3, win divided by stake, gives 3,219 not 3,081
can anyone explain? regards these "bf""e" means equvalent calculations,but don't know this formula

over 3

Hello,
I'm goin' to write some side thoughts here.
The short answer to your question is - if you try to extract some points indirectly via their published odds, you'll keep or even may accumulate the "vig" / "margin". And that approach to calculating some lines may be dangerous for your capital in my view, due to kept/accumulated negative EV.

Are you familiar with the Maths?
First, search on the Internet about the Poisson distribution.
That is basic for the initial educational ideas for building the probability of the goals scored in the match. Learn the idea behind the Poisson very well and build the probabilities for X, Y, or Z numbers of goals to be scored without the bookie's margin applied.
The next point is building various models, much closer to reality than the basic kind of the Poisson model, described on the Internet. Agencies' Stats departments are hard-working here. Various historical tests are required. Better models than Poisson are possible. Why not multivariate Poisson or Weibull distributions on some occasions and so on. I'm sure the agencies use automated models, based on AI and other advanced techniques.
For a better basic description of bivariate Weibull distribution, look out for a link published by the University of Salford, or other academic places.

Your own observations and deeper understanding of the soccer game are irreplaceable if you aim to be a winner in the long run.
For example, I participated in the soccer market South Korea K-Ligue 1, Suwon Bluewings - Ulsan Hyundai, held on 05.05.2022.
Due to some pre-analysis, I saw, that the published odds prior to the Kick-Off of that match were simply absurd. I spotted the possibility of huge profits, but I have restrained to open some positions here yet. Why? I always watch closely the Asian domestic agencies' odds, especially for the Asian events, because they must be most informed about them. I was shocked by their implied probabilities/odds prior to the match. 1X2, Goals Over/Under, and Asian Handicap odds were totally wrong in the agencies. There were two possibilities - they're right in terms of probability distribution here (most likely occurrence) or my model spotted some hypothetically valuable rare occurrence.
When in the first half I spotted many unusual accumulated cases for the Korean Soccer - a red card given, many substitutions, plus some changes in the game tactical approaches on the pitch on the fly, my positions were placed not only according to my predefined points but according to the new additionally spotted info as well, which influenced heavily to some computed probabilities.
In that Korean match, the preliminary calculated Poisson/Weibull etc were useless to some extent due to the circumstances on the pitch.

In short:
- avoid accumulating margin;
- try to build your own score probability models (the best choice) or modify well-known models, much closer to the reality and better applied to the reality than the basic simple model, described on the Internet;
- Fast reaction when new tactics are applied on the pitch.

if you profitable over 1000+ with your strategy, congrats you have found another loophole,but wasn't the question...

btw. the formula is correct,odds was wrong

Hello, sorry for the late opt-in, I'm very busy lately.
I'm glad that you've found what you've needed (exact formula etc) and wish you lots of profits.
And you're absolutely right in your post, the business became acceptable after many thousands of events. But not through a loophole at all, but through steady mental work and eliminating the errors via Socrates' principle.

Regarding your confusion about my previous post, let me explain.
The previous post contains side thoughts and aims to be useful for some players, reading here the thread, to avoid accumulating bookies' margin.
There is a possible approach:
- instead of inputting the bookies' reduced by margin odds in the exact formulas (that causes -EV by default), the player may compile her / his own probabilities by several known methods or improve their own models. For example, one of them is hinted here at blogs.salford.ac.uk/business-school/wp-content/uploads/sites/7/2016/09/paper.pdf
- when the probabilities are computed by the player, then have to convert into odds. The next possible step is to pursue + EV and seek to best available odds, in the time interval when the odds are aired from the agencies to the Kick-Off time (let's say, 5 to 7 days).
- The circumstances on the pitch or the coaching staff's decision may eliminate to a large extent these preliminary, Stats based calculations, or strengthen them. Described Korean domestic match is only one possible example. I mean, the player has to react fast, open new entries when they're triggered, or fast exits (take profit or stop loss) when necessary. Imagine a player participating in some markets in Ascot, making odds, putting positions, and spotting that a horse is suddenly changing its behavior in an unusual way.

@Arb12 - that’s a very nice article.
Do you use a bivariate weibull yourself?
I use a combination of independent negative binomial distribution and direct classification of events, and I wonder about the value of implementing such a model.

ex-hft wrote: ↑

Thu Jun 23, 2022 7:33 pm

@Arb12 - that’s a very nice article.
Do you use a bivariate weibull yourself?
I use a combination of independent negative binomial distribution and direct classification of events, and I wonder about the value of implementing such a model.

Hello again Ex-Hft,
My approach has some similarities to yours. Let me explain below.
I love science, and I'm grateful to the scientists and all creative minds in humankind's history (thinkers, philosophers, humanitarians, medics, engineers, etc), who shaped the contemporary civilization's achievements today as it is.
But I realize, that the pure science in some cases when it comes to random events, is more or less a conception in its purest kind. The application part of the theories in the world of random events is much more a kind of adventure.
My work in some defined by me cases includes modified (simplified) by me multivariate Poisson-like / or simplified multivariate Weibull-like approaches, as a start for building my own probabilities, but later everything is upgraded by my own observations. The pure Poisson is only acceptable for educational purposes to explain the idea of ​​the occurrence of 0,1,2 ....x...y...z number of events in a given interval, when some conditions are covered, for instance, independence of the events and so on. Do they cover all the required conditions 100 percent all the runtime?
In the Basketball game, that theory is on a pretty OK level, aiming for the understanding of the random distribution process. Not for wagering. For soccer, things are worsened in my view.
In short - pure academic theory as a ground is pretty OK, but the variables used in the formulas have to be your own know-how, built for many years. Later the formulas have to be influenced by additional quality factors, selected by you, and slightly modified. And year by year, diggin' into the specificity of a given kind of sport (i.e. the probability for X or Y, influenced by a successful tactical change by Pablo Laso in Basketball or an unsuccessful one from Mourinho in Soccer? The influence of a wet outfield in the cricket clash when Rahkeem Cornwall is goin' to be part of the batsmen pair? Or what is the influence of Rahkeem Cornwall's weight if he today is a part of the West Indies' bowling dept and if he bowls against a top-notch batsman like Virat Kohli in an International match? How the formula for computing some probabilities will change if you spotted or suspect a new approach by Frankie Dettori or Franny Norton? Or how the probabilities will change, if starting lineups are unusual - a different pitcher or batter, their Stats in the Baseball clash, and many more).
In short - the pure academics theory must be developed or simplified further by the player and Stats and the Data have to be part of your own variables and later the quality of the fundamental analysis must be processed. I mean the quality analysis to be transformed into data, and the noise in the process must be mitigated only by some sporting expertise by your side, nothing more.
Alongside the above, the risk and money management processes must be of the same quality and own conception otherwise you may raise the probability of failure.
In my practice very often some probabilities, computed by a simplified Weibull-like method were wiped when unusual starting lines-up were published, strange tactical decisions were made, or many many more. These cases taught me to reassess many things on-the-fly. The classification and weighting of various things and factors for no time, in pre-analysis or spotted on the pitch in in-play mode, and their quantitative equivalent is the real hidden value here.
The written above is part of my approach. Several projects have to be finished by me when I have much more time and I may upgrade some conceptions.
Do not use pure academic theories in the real conditions, upgrade them, use your own variables to load them into the formulas, reset some data and stats, use backtesting, and so on. It's very complex, indeed.

Yes, I think we agree.
I tend much more towards the fully automated trading / very broad end of the analysis, with much less emphasis on specific events (I don't trade manually) - except if they can teach me something broader which is applicable over many games.

In any case - I'm wondering: can you asses how much moving from a Poisson distribution to a Weibull increases your margin? (for example, the authors of the paper double their theoretical margin. Of course it's not achievable in practice, but it gives an idea of the strength of their claimed effect)

It's a very concrete question for me, as I'm wondering how to weight such a project vs. other opportunities (weather, lineups, additional data providers, etc. etc.).

ex-hft wrote: ↑

Thu Jun 23, 2022 10:13 pm

Yes, I think we agree.
I tend much more towards the fully automated trading / very broad end of the analysis, with much less emphasis on specific events (I don't trade manually) - except if they can teach me something broader which is applicable over many games.

In any case - I'm wondering: can you asses how much moving from a Poisson distribution to a Weibull increases your margin? (for example, the authors of the paper double their theoretical margin. Of course it's not achievable in practice, but it gives an idea of the strength of their claimed effect)

It's a very concrete question for me, as I'm wondering how to weight such a project vs. other opportunities (weather, lineups, additional data providers, etc. etc.).

@ Ex-Hft,
My reply to your two aspects in the previous post
1. If I understand you correctly, you pursue fully automated Technical plus Fundamental analysis. Unlike Technical analysis automation by some algorithms, I personally don't trust in the quality of all aspects of the fully automated quality analysis. My way is for all the aspects to be weighted by my mind and my experience. I believe your number of market interventions and volume is much bigger. Apparently, your way of market wagering in that case is more appropriate and more modern. I think that there are no experts, that are seeing the same aspects when they comment on non-quantitative and relative aspects of the game. There is not possible to exist the equal viewpoints on the fundamentals. Therefore the algorithm if it is created by the experts for fully automated quality analysis may vary in some details and may help you in some cases, or harm you in other cases.
2. I'm afraid that I can't give you direct strong advice on multivariate Poisson vs multivariate Weibull, but I've got a personal impression here. I'll explain. In my approach, as expressed in the previous post, I use in some cases my practical simplifications of both multivariate methods. They are only a basic point for computing some probabilities and odds projections by my side (100, divided by the computed probabilities for decimal odds). Further, in the time interval prior to Kick-Off and later in the game, additional info is incorporated over them - when the starting line-ups are published, when some events happen on the ground, substitutions, etc, etc.
To answer your question, I checked some things from my work in retrospect. The multivariate Weibull seems to be a little bit better for me, but this may be due to the peculiarities of my working style, or my variables. Better check deeply on the Internet, and later heavy testings, before investing something in one or other approach.
Wish you a lot of profits.

Thank you Arb12, and to you as well.

Let's break it down step by step.

Over 3.75 is between over 4.0 and over 3.5.

If the outcome is 3, it's a loss.
If the outcome is 4, it's a half win (O3.5) and a half push (O4.0)
If the outcome is 5, it's a full win.

So the profit in the middle case should equal half the profit from the full win. This means that half of the potential profit from the bets must be allocated towards either leg. In other words, the stakes on each leg should equal the inverse of the other leg.

For your example:

4.663-1 = 3.663 // O 3.0, potential net profit
2.458-1 = 1.458 // O 2.5, potential net profit
3.663+1.458 = 5.121 // total net profits
3.663/5.121 = 0.72 // O 3.0 share of net profits if equally weighted (NB! conceptually cleaner to divide by 2, if this seems unmathematical to you
1-0.72 = 0.28 // O 2.5 share of net profits
0.72*2.458 = 1.76 // Potential payout (gross profit) on the correctly weighted O 2.5 leg
0.28*4.663 = 1.33 // Ditto for O 3.0
1.33+1.76 = 3.086

Correct answer was 3.081, so that's just the margin of error.

hubdubber wrote: ↑

Mon Jun 20, 2022 8:40 am

if you profitable over 1000+ with your strategy, congrats you have found another loophole,but wasn't the question...

btw. the formula is correct,odds was wrong

No, your formula is only correct for U2.75, since there the half-push outcome is a loss, and so the potential loss amounts/stakes/risk amounts are what's to be distributed evenly. Another way to calculate that might be easier to remember is to take:
1/(1-1/4.663) = 1.273 (U 3.0)
1/(1-1/2.458) = 1.686 (U 2.5)
(1.273 + 1.686)/2 = 1.479 (U 2.75)
1/(1-1/1.479) = 3.086 (O 2.75)

ahh thanks for clarification