Elo rating system (my thoughts)

[Lardonio]

Re: Elo rating system (my thoughts)

@Lardonio As above quantitive or qualitative analisys is just two different methods of acomplishing the same thing. I don't see a reason to neglect the math heavy methods just because you don't like them. As you say basicly it is to try to automate the process. It would surprise me a lot if nobody have done that, they just havent published the results risking making the value disappear. At least thats what I think. I wouldnt look for them in the secret bookie bashing systems though :D Personally I do it because I enjoy the process, wether I actually find the holy grail is less important.

No, I agree, there is no reason to neglect the math heavy methods just because I don't like them. But what I don't like about them, at least in this thread, is that we've seen a lot of the math heavy stuff and none of the actual profitable betting stuff. I can make 500 threads posting formulas and theories but if it does not end up in a betting system or a betting strategy with set rules then it's a pointless mathematical wank in my opinion. As someone asked earlier in this thread - what is the point of this thread?

[Machine]

Re: Elo rating system (my thoughts)

I can make 500 threads posting formulas and theories but if it does not end up in a betting system or a betting strategy with set rules then it's a pointless mathematical wank in my opinion. As someone asked earlier in this thread - what is the point of this thread?

But you cant put up one thread telling people how to "value" bet , so whats the point in you posting? Some might find his thread interesting,some might want to learn about rating systems.(been there done it, but still find it interesting) To come on and call it mathematical wank is pathetic. Wouldn't your time be better spent writing "Lardonio's How I value bet consistently" thread

[ELO Evangelist]

Re: Elo rating system (my thoughts)

No, I agree, there is no reason to neglect the math heavy methods just because I don't like them. But what I don't like about them, at least in this thread, is that we've seen a lot of the math heavy stuff and none of the actual profitable betting stuff. I can make 500 threads posting formulas and theories but if it does not end up in a betting system or a betting strategy with set rules then it's a pointless mathematical wank in my opinion. As someone asked earlier in this thread - what is the point of this thread?

The answer to your questions lie at the end of a years work developing a complex mathematical model. If you are not prepared to ask the question then you will never know the answer.

[Lardonio]

Re: Elo rating system (my thoughts)

Anyway enough of this' date=' this is ELO's thread about elo ratings not the 'mythical value bet'[/quote']

[Machine]

Re: Elo rating system (my thoughts)

It seems impossible to get you to understand what value betting is, so I suggest you take your own advice and focus on the ELO ratings.

I must know what value betting in sport is because for the last 22 years when I have ask value bettors what it is and how its done I keep getting the same answers. rolling dice :rollin flipping coins :rollin dealing cards :rollin If I backed the bet 100 times Id win (thats a classic):rollin In the long run (how long is this long run?):rollin I think the fav will win but Im looking for a value bet in the race (another classic):rollin Then it usualy ends up "you dont understand value" :ok I fully understand what value is. My problem is with people who say they're value bettors or talk about it yet cannot tell me how they do it consistently At least with ELO's ratings he can say team x has a higher rating than team y and then decide whether he wants to back it. Rather than team x should win but Im backing team y because I have produced a mythical figure. :eek

[Lardonio]

Re: Elo rating system (my thoughts) It seems impossible to get you to understand what value betting is, so I suggest you take your own advice and focus on the ELO ratings.

[carlos50]

Re: Elo rating system (my thoughts)

I fully understand what value is.

Could you explain it to us then?

[Machine]

Re: Elo rating system (my thoughts)

Could you explain it to us then?

Of course Value can only be correctly obtained when the true percentage of all possible outcomes is known, then and only then can you say that the %age payout is greater or less than than the true percentage. Even then it is possible to go into negative equity due to not having sufficient funds. Now Lardino instead of repeating yourself expose to the world how you consistently value bet in sports betting, or may I suggest its you who do not understand the concept of value in sports betting.

[Lardonio]

Re: Elo rating system (my thoughts) Hey monkey - I've never claimed to consistently value bet anything and if I decide to publish my betting methods on this forum it will certainly not be aimed at proving a point to you, as after I push 'Post quick reply' under this message field I'm not wasting another second on reading your posts and your pointless stabs at a topic you do not fully understand. I hope we can get this post back on track and learn how to bet profitably with ELO ratings.

[gloryhunters]

Re: Elo rating system (my thoughts) You don’t need to know all possible outcomes to have a value bet. In sport you don’t know the true probability, but estimate it using some system that you have developed and then compare to the price on offer.

On a single bet, the result won’t matter as you had value in the bet. The same way as tossing a coin, if somebody offered you 4/1 on heads and it came down tails, you still took a value bet.

The big question is how do you know if your system estimates the outcome accurately enough. This can be done through looking at a large sample size and test if you system significantly performers better than expected. The size of the sample will depend on the odds and how much value you are trying to measure, but there are statistical tools to work this out.

If you don’t monitor your results over a long period of time you’ll never know if your system actual works.

As far as I can work out, what is being proposed here is exactly the same. It ranks the team then applies a regression to estimate the probability of a win and then see if you get value.

Finally, looking for value is not the same as thinking something will win. For a basic example, rolling a dice and somebody says you can either have the numbers 1 to 5 at 1/20 or number 6 at 15/1. The value is betting on number 6, whereas you expect the result of the roll dice to be a number 1 to 5.

[Machine]

Re: Elo rating system (my thoughts)

You don’t need to know all possible outcomes to have a value bet. In sport you don’t know the true probability' date= but estimate it using some system that you have developed and then compare to the price on offer.

On a single bet, the result won’t matter as you had value in the bet. The same way as tossing a coin, if somebody offered you 4/1 on heads and it came down tails, you still took a value bet.

The big question is how do you know if your system estimates the outcome accurately enough. This can be done through looking at a large sample size and test if you system significantly performers better than expected. The size of the sample will depend on the odds and how much value you are trying to measure, but there are statistical tools to work this out.

If you don’t monitor your results over a long period of time you’ll never know if your system actual works.

As far as I can work out, what is being proposed here is exactly the same. It ranks the team then applies a regression to estimate the probability of a win and then see if you get value.

Finally, looking for value is not the same as thinking something will win. For a basic example, rolling a dice and somebody says you can either have the numbers 1 to 5 at 1/20 or number 6 at 15/1. The value is betting on number 6, whereas you expect the result of the roll dice to be a number 1 to 5.

SIGH :eyes

[Machine]

Re: Elo rating system (my thoughts)

Hey monkey - I've never claimed to consistently value bet anything and if I decide to publish my betting methods on this forum it will certainly not be aimed at proving a point to you, as after I push 'Post quick reply' under this message field I'm not wasting another second on reading your posts and your pointless stabs at a topic you do not fully understand. I hope we can get this post back on track and learn how to bet profitably with ELO ratings.

:zzz thats it keep your holy grail under wraps just talk about it but cant prove it all hot air

[Vilamoura04]

Re: Elo rating system (my thoughts)

I think hardly anyone is using ELO based models. And even if they are I wonder how many of them know how accurate their algorithm is compared to the others.

I've obviously misinterpreted but I thought your opening post said millions of Internet users were using your system.

[ELO Evangelist]

Re: Elo rating system (my thoughts)

I've obviously misinterpreted but I thought your opening post said millions of Internet users were using your system.

I see what you mean. I contacted my old uni. maths professor for a copy of the FORTRAN code and he told me who was now using it and where it was now being used. Yes it is probably being used by millions of people a day but they have no idea what backend is driving the website.

[Machine]

Re: Elo rating system (my thoughts) Have you crossed checked using virgin data? ie a data set not used in your predictive model

[ELO Evangelist]

Re: Elo rating system (my thoughts) On the subject of value if you ask one hundred people for a definition you will get one hundred different explanations. It always seems to cause disagreements on betting forums, I guess it is something that is internalised. When my calculations are done I am left with a figure that tells me my edge over the market price vs the model's predicted price. The next step in the process then takes place. My pet hate are punters (most of my friends included) who do not know what their edge over the market is. They will place a bet because their head to head system tells them that team A has won more times than team B therefore they will bet team A, but can not tell you what the bet's positive expectation is.

[Machine]

Re: Elo rating system (my thoughts)

When my calculations are done I am left with a figure that tells me my edge over the market price vs the model's predicted price.

If you are using market price vs the model's predicted price I would take a step back and have a rethink before you consider placing any money whatsoever. Run your model thhrough virgin data first and you will see why.

[Vilamoura04]

Re: Elo rating system (my thoughts)

I see what you mean. I contacted my old uni. maths professor for a copy of the FORTRAN code and he told me who was now using it and where it was now being used. Yes it is probably being used by millions of people a day but they have no idea what backend is driving the website.

I'll bite. What website are we talking about?

[protop]

Re: Elo rating system (my thoughts)

Home = 0.9429 Away = 0.9428 Draw = 0.6927 Makes your mouth water doesn't it. Even my maths mentors are impressed they told me to aim for 0.7 across the board but the Home/Away figures far surpassed this.

Hi Can I ask how you measured your probabilities in terms of r-squared numbers Thanks Protop

[ELO Evangelist]

Re: Elo rating system (my thoughts)

Hi Can I ask how you measured your probabilities in terms of r-squared numbers Protop

Sure, I assume you you are referring to the X & Y terms of the regression. The X term was the (ELO algorithm) rating difference between the two competing teams and the Y term was the observed outcome as a % (win, loose, draw). 27,000 results were grouped into 150 data points that were used in the plot.

[ELO Evangelist]

Re: Elo rating system (my thoughts)

If you are using market price vs the model's predicted price I would take a step back and have a rethink before you consider placing any money whatsoever. Run your model thhrough virgin data first and you will see why.

Not quite sure what you mean by testing against virgin data. Do you have a specific way this can be done with ELO algorithms in mind or do you mean it is a good idea to save some of your training data for testing? As a simplified explanation I train my ELO algorithm with 10 - 60 years worth of results data then test in against 1 - 10 years worth of observed odds data. There could be a slim argument that the results and the odds are self referencing but I don't think so as the odds are frozen at kick off and so have no influence on the result.

[protop]

Re: Elo rating system (my thoughts)

Sure' date=' I assume you you are referring to the X & Y terms of the regression. The X term was the (ELO algorithm) rating difference between the two competing teams and the Y term was the observed outcome as a % (win, loose, draw). 27,000 results were grouped into 150 data points that were used in the plot.[/quote'] Hi Thanks for your reply. When you say that you have 150 data points, I assume they range from say a high negative number to say a high positive number, -75 to 75 for example. Did you then apply these to your R squared formula to get the appropriate probability. If so how many results have you applied the optimised ratings against and do they continue to perform as expected. It is an area of interest of mine. Thanks Protop

[Machine]

Re: Elo rating system (my thoughts)

Not quite sure what you mean by testing against virgin data. Do you have a specific way this can be done with ELO algorithms in mind or do you mean it is a good idea to save some of your training data for testing?

When I say virgin data I mean data that has not been used in your predictive model. I can see that your falling into the trap of assuming that if your models predictive price has a +edge against the bookies odds all you have to do is back at that price or higher and you will make a profit (if only it was so easy)

[Gingertipster]

Re: Elo rating system (my thoughts) Interesting stuff. However.... Any punter trying to find the value in any sport needs to take in to account all sorts of things. How does an ELO system take in to account recent form? If ELO shows Man U to be the best side over a period of time (say a season). What if Wayne Rooney is in poor form? What if he plays better with Giggs in the team, and he's missing? What if Rooney isn't playing? What if the team is scoring more than usual / less than usual? How does ELO take in to account the new signings? Or a new manager? What if 3 of the usual back 4 are missing? Is the teams head to head taken in to account? How is confidence (at the point of kick off) taken in to account? What if they've lost their last three matches? If you take the last question, there will be times when those 3 matches mean a great deal, other times (e.g. if they've lost Cup games putting out a second 11). There will be times when something has a great influence on the odds, other times very little influence. Systems find it hard to control these things. There are so many more things that effect a team's performance than how good they've been over a season (or other period of time). All the above and more, contribute to how a team is likely to play and so effect if a team is value (in odds/percentages terms). Value will always be subjective. A System, or rather Method can no doubt help. But VALUE comes down to opinion at the end of the day. IMO the best way of betting is to use your knowledge of the sport to work out the percentage chance of each outcome, convert to odds and back the one/ones who are VALUE. If you are good at evaluating the form, you'll win overall. If you're poor at evaluating form, you'll lose overall.

[ELO Evangelist]

Re: Elo rating system (my thoughts)

When you say that you have 150 data points, I assume they range from say a high negative number to say a high positive number, -75 to 75 for example. Did you then apply these to your R squared formula to get the appropriate probability. If so how many results have you applied the optimised ratings against and do they continue to perform as expected.

When the rating difference is +ve the home team has the higher rating and when the difference is -ve the away team has the higher rating. I do not combine these two groups into the same regression. Instead I separate them. One data set when the home team has the rating advantage and one data set where the away team has the advantage. To get the probability of win, loose, draw rating differences are grouped. So all fixtures where the rating difference was between 0 and 0.1 are analysed to create three data points(win, loose, draw). For example if there are 200 results in the 0 - 0.099 rating difference range then the three data points are derived from how many of those fixtures resulted in a win, loss or draw. The same is then done for 0.1 - 0.199 etc.

[ELO Evangelist]

Re: Elo rating system (my thoughts)

When I say virgin data I mean data that has not been used in your predictive model.

You can not keep test data away from your training data with an ELO ranking algorithm because it relies on tracking the incremental chances in a teams performance through time. If you were to keep one years results away from the ranking algorithm then it will create and inaccurate assessment of the team. What I do is after the algorithm has given me a prediction of the teams ability through time I then go back in time and see what odds were on offer compared to what the model predicts what the odds should have been. The historic odds are not part of the training data only the score/result is.

[ELO Evangelist]

Re: Elo rating system (my thoughts)

Interesting stuff. However.... Any punter trying to find the value in any sport needs to take in to account all sorts of things. How does an ELO system take in to account recent form?

This is one of the strengths of an ELO algorithm because it only gives you what the teams current ability is. It slowly tracks through time the small changes in a teams performance and updates the teams ability match by match. On the other hand one of the weaknesses of an ELO algorithm is that it is strictly a form based assessment of the teams ability and can not take into account the fact that a team's top striker is out for the rest of the season. The algorithm will detect a drop in the team's form and lower it's rating. I agree with you that knowledge of the sport you are betting in can be a great advantage if that knowledge is applied without bias.

[muppet77]

Re: Elo rating system (my thoughts) My problem with elo type ratings is that you have to convert them into odds (unless you bet on spreads). By doing this you are using a model of a model, that is firstly the elo model to get a rating, then secondly a model to convert the values to prices. How do you overcome this? What technique do you use?

[salford_student]

Re: Elo rating system (my thoughts)

My problem with elo type ratings is that you have to convert them into odds (unless you bet on spreads). By doing this you are using a model of a model, that is firstly the elo model to get a rating, then secondly a model to convert the values to prices. How do you overcome this? What technique do you use?

Muppet, the strength of ELO ratings is that they can be easily converted into probabilities which are then easily converted to odds. http://en.wikipedia.org/wiki/Elo_rating_system

[salford_student]

Re: Elo rating system (my thoughts) ELO evanagelist, if your rating system is so good why not enter it in this competition so we can see how it performs. http://kaggle.com/chess