ELO & Form Ratings System / New Approach

[LegendMon]

Hi Guys,

I've been working on very arbitary rating systems in the past, with mixed results, until discovering ELO ratings. Over the last few months I have been working on developing my own improved rating system which seems to be working more solidly for the time being.

With all the reading that I was doing, I found that it was common that each teams ELO rating continues from season to season. I'm not entirely sure what the reasoning behind that is, or how accurate that can be -- especially when you take into account promotion and relegation. My personal view is that each team may be very active in pre-season, have a new manager, have a new billionaire owner. This could skew the ratings massively in favour (or against) certain teams. Take a league like the Eredivise, or Allsvenskan, where players move on with regularity. A large team can easily be decimated, and a small team can improve with exciting youngsters. I like to see that a new season as a fresh start, and their results from ten seasons ago don't matter. 

For that reason, my method makes each team start on the same amount of points at the beginning of each season. I then see each league as a long poker tournament where the home team and away team both post their respective big blind (7%) and small blind (5%) into the pot, respectively. The ratings for each team then grow/reduce as usual (I use a goal supremacy system to divide the pot). To strengthen the short term success, it take into account the overall progress combined with the last five home or away games, in order to get a final percentage for a home win, draw and away win. This system, therefore, works best after 10 games so recent form can take effect.

Here's an example from last weekend Swedish Allsvenskan:

Djurgården: 42,93% / Draw: 21,09% / Orebro: 35,97%

Final result was 3-2 to the home team. According to the odds before the match (1.56), Djurgården were not value.

Any thoughts?

 

Edited October 6, 2016 by LegendMon
Clicked submit too early..

[Sir Puntalot]

Hi LegendMon,

Are you referring to our ELO ratings or just in general?  

Mine have always start fresh at the start of the season also. I'd have to disagree about waiting 10 games, I think you can see things as early as 1 or 2 games, this is when they are most dangerous in my opinion. Once 10 games are played, a clear pattern in how each team plays is formed and the bookies are absolutely over that after 10 games.

My ratings have a last 4 rating and last 2 rating and when we have the new design up and live, I intend to extend the ratings to cover more leagues.  

[LegendMon]

Hi Paul,

Wow, wasn't expecting an almost instant reply!

I mean ratings in general. To be honest, I haven't seen your ELO ratings (blush) -- do you have a link and/or thread where I can see these, just so I can do some additional reading? It's interesting and relevant that you also start fresh at the beginning of each season, so I'm super intrigued to get more info. Apologies if that's a newbie question

Thank you for the comment, and also disagreeing with me. That's what I was hoping someone would do! I guess I could implement a similar last 4/last 2 rating, but this is all relatively new to me. Did you find those numbers to be a better indication of "form"?

I usually bet the double chance, so I look for games where the draw is least likely (slightly better odds). Too many times, I've been stung by a small team grabbing a 1-0 win!

[Sir Puntalot]

Apologies for the poor design (big improvement coming very soon!), but here they are https://www.punterslounge.com/football/ratings

You can access using the navigation menu above too.  



I'll reply with more later, but basically I brought last 2 average into it also because I can then see ok - Team A is 20pts ahead on last 4 averages, but it's actually 15pts behind on the last 2 average. It's not mega often that happens, but it does virtually every week in one of the leagues if not more.

[LegendMon]

Aha, excellent stuff.

Understandably you won't disclose what parameters you use to get those ratings, but I can see how it works with the ratings like you explained above. I had a quick comparison with the next round Championship matches; some match mine, some don't. I have a "rating" and also the 1X2 breakdown. Let's take Cardiff vs. Bristol as an example. I get the following percentages:

Cardiff: 23,41% / Draw: 17,56 / Bristol: 59,02%

Bristol obviously look to be great value here (in both cases), and I'd probably be looking at the 12 double chance.

If I had time then I'd definitely look into adding more parameters into my own system. In your experience, is the more parameters, the better, to find out how a team performed -- rather than looking into the goal supremancy of a result? I intend to keep a log of the remaining matches this season, to see if this works in the long run. I probably won't post all my selections on a weekly basis though, until I'm confident in them. But it's certainly interesting stuff!

[Sir Puntalot]

I wouldn't worry about it, just post away, easier to keep a record.  

As for ratings methodology - keep it simple and logical, that's it. Mine hasn't changed in 12 years.

[LegendMon]

Thanks for the advice and feedack, Paul. Much appreciated!  I'll be back and post some selections after the international break -- gives me some time to get some other leagues into my database too.

[LegendMon]

I only have the one league (Sweden Allsvenskan) ready to show potential results, so far, based on the rules I set. Here are all the games in the next round of fixtures, with rounded up percentages for each result. I also have a "rating" for each game, which is no way related to Paul's work.

[LegendMon]

And for the Championship, now that I had time to put the system together. Note: the ratings are unique for each league. I need to work on developing a common standard acorss all leagues.

[LegendMon]

By my calculations that was 14/20 correct; not too bad. I didn't take down any odds or anything, which I probably should do next time, unless anyone knows where to find historical double chance odds? For the time being I just wanted to see how it worked! Will add Premier League into the mix for this weekend. I like how some of the results don't always go the way of the "favourite", for example, Derby vs. Leeds, and the way I do it covers the "underdog" to win too (in most cases). I really try to avoid the draw where possible

[world]

Hi LegendMon,

Interesting ratings, I also like double chances. I'd like to follow your ratings with this Saturday's English Premier league and English Championship matches. Could you please publish it here?

Edited October 19, 2016 by world

[world]

On Thu Oct 06 2016 18:17:42 GMT+0300, Paul Ross said:

I wouldn't worry about it, just post away, easier to keep a record.  

As for ratings methodology - keep it simple and logical, that's it. Mine hasn't changed in 12 years.

Hi Paul,

I know your ratings are very good to find value bets. I was wondering how I can use them for unibet's supertoto type of bets, where it's possible to choose singles, doubles, triples. Also, is there a way to convert your ratings to 1X2 percentages?

[world]

On Thu Oct 06 2016 18:17:42 GMT+0300, Paul Ross said:

I wouldn't worry about it, just post away, easier to keep a record.  

As for ratings methodology - keep it simple and logical, that's it. Mine hasn't changed in 12 years.

Hi Paul,

I know your ratings are very good to find value bets. I was wondering how I can use them for unibet's supertoto type of bets, where it's possible to choose singles, doubles, triples. Also, is there a way to convert your ratings to 1X2 percentages?

[LegendMon]

16 hours ago, world said:

Hi LegendMon,

Interesting ratings, I also like double chances. I'd like to follow your ratings with this Saturday's English Premier league and English Championship matches. Could you please publish it here?

Hi @world, I absolutely can post them here for you to see. Like I said earlier, these are still kind of work in progress and each "rating" is league-specific at the moment. You will still see the percentages the system throws out for 1X2 and double chance, though. I aim to post them on Friday at some stage. Hope that helps!

[world]

Thanks

[LegendMon]

Hello again, @world and everyone else!

Here is the EPL and Championship games, now that I'm finished updating the stats. Remember that the rating between leagues are not on the same scale (each competition is a different "tournament" if I keep my poker analogy going!). I have omitted the Swedish Allsvenskan since I guess I'm the only one interested in that, but if people want it, then no problem.

EPL ratings, are probably close, but ideally I'd want teams to have played 10 games before suggesting bets, so take them with a pinch of salt.

It's interesting that there are some home teams are favourites, but the game suggests a (slightly) stronger away team - for example, the West Ham vs. Sunderland game.

I'm always open for discussions!

[world]

1 hour ago, LegendMon said:

Hello again, @world and everyone else!

Here is the EPL and Championship games, now that I'm finished updating the stats. Remember that the rating between leagues are not on the same scale (each competition is a different "tournament" if I keep my poker analogy going!). I have omitted the Swedish Allsvenskan since I guess I'm the only one interested in that, but if people want it, then no problem.

EPL ratings, are probably close, but ideally I'd want teams to have played 10 games before suggesting bets, so take them with a pinch of salt.

It's interesting that there are some home teams are favourites, but the game suggests a (slightly) stronger away team - for example, the West Ham vs. Sunderland game.

I'm always open for discussions!

Thanks! As I'm from Finland I also follow Swedish Allsvenskan:). Can you please explain a bit what those numbers in Rating column means, how to read them?

[Sir Puntalot]

On 10/19/2016, 4:45:52, world said:

Hi Paul,

I know your ratings are very good to find value bets. I was wondering how I can use them for unibet's supertoto type of bets, where it's possible to choose singles, doubles, triples. Also, is there a way to convert your ratings to 1X2 percentages?

@worldThey're good to find all kinds of bets because they're so open to interpretation. Like any ratings, they can only be a guide but they've been consistently good for over 10 years with no change to the method.

Converting ratings to percentages - interesting call, I'd have to think about how to go about that. 

[world]

On 21.10.2016 klo, LegendMon said:

Hello again, @world and everyone else!

Here is the EPL and Championship games, now that I'm finished updating the stats. Remember that the rating between leagues are not on the same scale (each competition is a different "tournament" if I keep my poker analogy going!). I have omitted the Swedish Allsvenskan since I guess I'm the only one interested in that, but if people want it, then no problem.

EPL ratings, are probably close, but ideally I'd want teams to have played 10 games before suggesting bets, so take them with a pinch of salt.

It's interesting that there are some home teams are favourites, but the game suggests a (slightly) stronger away team - for example, the West Ham vs. Sunderland game.

I'm always open for discussions!

HI LegendMon, It was my pleasure to follow your ratings. In Premiere League: 8/10 (80%) correct and in Championship: 10/12 (83%) correct...Excellent!!! I hope you'll post your ratings also in future.

[world]

4 hours ago, Sir Puntalot said:

@worldThey're good to find all kinds of bets because they're so open to interpretation. Like any ratings, they can only be a guide but they've been consistently good for over 10 years with no change to the method.

Converting ratings to percentages - interesting call, I'd have to think about how to go about that. 

Thanks, Paul. Yeah, I found also your ratings can be interpreted in many other ways. Converting %...I was thinking if we know the max ratings a home/away team can score then we could convert them easily:), also knowing % it'd be easier to find the value!

[Sir Puntalot]

That's just it, there is no maximum, so it needs some thought. 

I'll have a think, but we're in the middle of a complete redesign (ratings look superb in the new style) too, so I might be a while. 

[LegendMon]

Hi @world, thanks so much for the kind comment! I'd probably take those kind of hit frequencies every week, as long as I got value, of course. 

To answer your question about the rating column, and I'll try be as brief as possible here: each team started on the same amount of points, and with each game played, depending on the result, their points are updated. I then calculate the average points overall and the average from the last 5 home and away games. From there it's a simple formula to get the values in that column. A value of 0 suggests a close game, whereas a large minus number could be a stronger away win, and vice versa. The 1X2 percentages are calculated from the total points of each team, cross-calculated with their actual results.

I will definitely post more games, and also looking to add England League One and Two soon!

As for Sweden Allsvenskan, Malmö had a terrible game on Saturday! EDIT: In fact, technically it was the biggest shock for the entire season, if we're going by ratings.

Edited October 24, 2016 by LegendMon
Added info.

[world]

11 hours ago, LegendMon said:

Hi @world, thanks so much for the kind comment! I'd probably take those kind of hit frequencies every week, as long as I got value, of course. 

To answer your question about the rating column, and I'll try be as brief as possible here: each team started on the same amount of points, and with each game played, depending on the result, their points are updated. I then calculate the average points overall and the average from the last 5 home and away games. From there it's a simple formula to get the values in that column. A value of 0 suggests a close game, whereas a large minus number could be a stronger away win, and vice versa. The 1X2 percentages are calculated from the total points of each team, cross-calculated with their actual results.

I will definitely post more games, and also looking to add England League One and Two soon!

As for Sweden Allsvenskan, Malmö had a terrible game on Saturday! EDIT: In fact, technically it was the biggest shock for the entire season, if we're going by ratings.

Hello @LegendMon, Thanks for the explanation, I understand now. Yeah, who would've thought that Malmö would lost that match at home, I guess no ratings can predict that. I followed a very nice preview (in English) about Allsvenskan 2016: https://www.reddit.com/r/soccer/comments/4d0rei/allsvenskan_starts_today_heres_a_guide_to_the/ As it says: Allsvenskan is one of the most unpredictable leagues in Europe, as proved last year when IFK Norrköping, a team most people thought would end up in the bottom half, managed to win the entire thing with an impressive points tally of 66. Same thing actually happened with Finland's Veikkausliiga this year, a club ( IFK Mariehamn ) from an small island won the league and surprised everybody.

I'll be very happy if you post your ratings here for every Saturday (atleast Eng. Premiere, Championship and Allsvenskan).

Edited October 24, 2016 by world

[aftershock9]

Hey mate, love your work with your ratings. May I ask a question? In the bet column you suggest a strategy to bet for a particular match. If you look at your suggestions for the past EPL weekend, your strike rate is 80% but those double chance bets are short priced odds (1.1 - 1.3) and even with a 80% strike, you will end up losing with a flat staking approach. How do you approach your betting strategy with the ratings?

[LegendMon]

12 hours ago, world said:

I followed a very nice preview (in English) about Allsvenskan 2016.

I'll be very happy if you post your ratings here for every Saturday (atleast Eng. Premiere, Championship and Allsvenskan).

Hi @world, thanks for that link I'll definitely check it out since I live in Sweden! Ratings to come every week where possible, probably before the weekend because I believe some Swedish games are played on a Friday (but don't quote me on that!).

[LegendMon]

11 hours ago, aftershock9 said:

How do you approach your betting strategy with the ratings?

Hi @aftershock9, thanks for the comment. That's actually the next step, I guess. I'm not placing any money on these ratings just yet; it's more of seeing what works and what doesn't at this stage. Without much thought or theory behind it, I would say that I could adopt a confidence level kind of thing, where the stake is adjusted to the rating. Or, I might decide to bet only the ones that show value. I'll give it some thought and get back to you

[Sir Puntalot]

I'd be careful with minor leagues guys. I've always found that the lower standard you go, the more things runs into trouble - it applies to women too.  

[LegendMon]

Can't argue with Sir Puntalot!

Anyway, as requested by @world, here are the Swedish Allsvenskan for the next round of matches (tomorrow and Thursday).

Taking into account the previous comment about betting strategy, there's only three games that represent any double chance value (based on Bet365 odds at least). They are highlighted in the image.

EDIT: I will start posting real "bets" over the next week or so, so we can gauge how successful this could be.

Edited October 25, 2016 by LegendMon

[world]

@LegendMon, Thanks a lot for the Allsvenskan ratings. At the moment I bet only on unibet supertoto or Swedish Stryktipset type of bets where I can have very small stakes and choose freely singles/doubles/triples, so finding value bet is not the priority. That's why I like your ratings with double chances.

@Sir Puntalot...oh man, that's so true.

 

Edited October 25, 2016 by world

[gogolitalia]

Great Idea but be careful with Sweden since there are only 3 matches left so many teams will play for nothing and other will play for to stay in the league. Kalmar-Gefle is one of those matches where Kalmar wont get into Euro league while Gefle can still avoid direct relegation. I think you need to take these circumstances when deciding the matches. Other than that I follow your posts