Challenge - Make our own model - over/unders goals

[Semibluff]

Ive often wondered about how the exact models are made to compile the over unders market for exact goals in a certain game.
What exact variables they use to compile this?
Is it an exploitable market?

It is normally very liquid close to kick off and yet compared to match odds its very hard to determine how right or wrong the prices are.

Following on from a question posed to john in his thread, and from examining Straddles model which he uses I was wondering is there a way to see if as a group, picking up on each others ideas or opinions we can get closer to an exact model to use for the games.

What i propose is that we pick a fixture 2 weeks away, and anyone interested in having a go at this would do their own "model" and adjust it according to their own weights for each variables.

The game i would like to pick is;

Man City @ home to Wolves. 15th January.

Variables i can think of off the top of my head:

goals average for the month of jan
Team averages
Manager goal averages
2/3 important players for defensive performance
2/3 important players for attacking performance
Weather
Motivation
Suspensions
Formations

The fun bit is assigning an exact weight and trying to find out how accurate you may or may not be.
The game picked has a long standing manager with an unchanged team, aswell as a manager with different records and alot of attacking players. Should be interesting.

Anyone interested should announce so here. And then whilst defining their model they can watch the next few games in the league to help guide them. (the over under 2.5 games market can be found on bf)

Then as soon as possible, before the markets are properly liquid, we should post what price we would make it (those posting earlier would be a lot ballsier!)

When we are finished - we can mail each other how we deduced this price, and see if theres any Major flaws in our thinking that we could adjust.

I will be doing this in the new year myself regardless, but i also think for those into their betting this could be alot of fun too.

If your not interested in this, or would rather keep your ideas to yourself, then thats 100% ok, sorry for boring you!

Otherwise i think it could be a lot of fun. and with only game to work on shouldn't be too time consuming (challenge extended to those who wouldn't normally bet!)

Tommy Gunne et all, get involved!!!!

[ShipIt]

I like ur idea but why pick a no brainer market to compile. Pretty sure any mug punter I know would guess and wouldnt be 20pips out.
If its football u want to concentrate on I would suggest sticking to match odds.
Also If a few of us agree we have to post/mail our prices a week before and must lay them prices up to a certain amount.

[Mullicker]

goals average for the month of jan
Team averages
Manager goal averages
2/3 important players for defensive performance
2/3 important players for attacking performance
Weather
Motivation
Suspensions
Formations

I would scrap:
Historical data for Jan
Formations
Weather
Manager and goal averages


The important attacking and defensive players could be useful and that could escape most peoples attention.

The main stats i would use:
-avg home goals scored
-avg home goals conceded
-avg away goals scored
-avg away goals conceded
-Results at home vs similar teams
-Results away vs similar teams

[Mullicker]

The fun bit is assigning an exact weight and trying to find out how accurate you may or may not be.

How do we know which model is more accurate?

How would you test a model you have developed. I guess you would need about 10 seasons of data that you can be entered into your model, and record the result of each game vs your model. Then mess around with your model to see if you can optimize. I'm not even sure this would give you a great model, and you would probably need a programming and statistical background to make some attempt at this.

[Semibluff]

I like ur idea but why pick a no brainer market to compile. Pretty sure any mug punter I know would guess and wouldnt be 20pips out.
If its football u want to concentrate on I would suggest sticking to match odds.
Also If a few of us agree we have to post/mail our prices a week before and must lay them prices up to a certain amount.

I think its far from a no brainer market to compile. I think its a market that gets punted in a lot by the general gambler who doesnt really have a clue what price it should be, or what exactly they should be looking for. Its something im going to look at myself to try and see how exploitable it is as im aware the bigger more professional punters have already identified it as a good market.

The mug punter being 20 pips out, the diff from 1.9 to 1.75 is were your going to get your margin. It leave you the recommended % margin for error and give you a valuable bet!

Im already doing some betting myself on match odds - but with the prem being so heavily liquid its very hard to find too many errors. If anyone has questions about how im doing the match odds, pm me.

[Semibluff]

I would scrap:
Historical data for Jan
Formations
Weather
Manager and goal averages


The important attacking and defensive players could be useful and that could escape most peoples attention.

The main stats i would use:
-avg home goals scored
-avg home goals conceded
-avg away goals scored
-avg away goals conceded
-Results at home vs similar teams
-Results away vs similar teams

This is the beauty of it. Each person could decide themselves whats relevant and whats not.
Agree about the above, also important. Just goes to show the lengths you could go to to get the exact information!

[Semibluff]

How do we know which model is more accurate?

How would you test a model you have developed. I guess you would need about 10 seasons of data that you can be entered into your model, and record the result of each game vs your model. Then mess around with your model to see if you can optimize. I'm not even sure this would give you a great model, and you would probably need a programming and statistical background to make some attempt at this.

Well for the sake of this we wont know too accurately until the match odds. Bearing in mind that pinnacle and sbo have their lines up about a week earlier than the game, im not sure how to have it so that we can "bet on it" who find out who's wrong or right.

The normal process with something like this would be to define the model loosely, now apply it to a full fixtures games (this would involve compiling the same data for all of the teams, stats etc)
Through trial and error youll eventually find yourself agreeing with most of the market prices. Its now youll know you've a happy medium of the variables, their strengths and confidence in your own model to start weighing in behind in bets.

Once you adjust accordingly for short term goals scored and motivational aspects then it should work for there on in.

[Semibluff]

Also - when compiling a teams average goals, When referring to a 5-0 win, i recommend only counting 2 goals, then .5 of a goal for the next, and .25 goals for anything after inclusive. so this game would be 2.75 goals

[digiman]

Also - when compiling a teams average goals, When referring to a 5-0 win, i recommend only counting 2 goals, then .5 of a goal for the next, and .25 goals for anything after inclusive. so this game would be 2.75 goals

Why would you do this?

I don't sports bet at all really but like your idea.

[Semibluff]

It something i was told to make sure i do by someone who know's this sort of thing and when i thought about it it really made sense. Mainly because it could really mess up your average

Lets take newcastle for example - they dont score too much, yet had 5-1 vs sunderland and 6-0 vs villa

If you allowed these games stand on merit then you would have a pretty high goal average for Newc whereas the true figure is a lot lower.

When teams are beaten a number of things happen
They throw the CB up front
They drop the heads
They sub off big players
They play silly 3/4 man upfront formations that dont fit their team

They affectively panic and are more suspect to conceding more goals. So these extra goals shouldnt be allowed have too much merit in the greater scheme

Even if you dont sports bet, could still be a bit of fun doing this - is only a numbers game really

[Mullicker]

Also - when compiling a teams average goals, When referring to a 5-0 win, i recommend only counting 2 goals, then .5 of a goal for the next, and .25 goals for anything after inclusive. so this game would be 2.75 goals

I kinda like the logic behind this, sort of to allow for freak results i presume.

[Bubbleking]

The model I use for GAA betting.

1. the history of the teams involved (last 10 games with with different weightings given to league, champo and cups etc)

2. time of the year (includes weather such as wind, rain etc etc)

3. the venue (different size pitches means more/less scores)

4. the ref (whistle happy equals more frees and more scores)

5. current form

6. Inside info

7. manager tendencies

that's all I can think of I prob look for other things too that Im not aware of. It can prob be used for soccer too. I have a pretty substantial edge in the GAA market but it isnt very liquid which is a pity

[Mullicker]

The bookies are very stingy with their prices for the GAA. Never done that much punting on the GAA but imagine i would do pretty well on local club championship games in Waterford, but don't know how many bookies offer this. Inside knowledge would be massive when it comes to club games.

[Bubbleking]

The bookies are very stingy with their prices for the GAA. Never done that much punting on the GAA but imagine i would do pretty well on local club championship games in Waterford, but don't know how many bookies offer this. Inside knowledge would be massive when it comes to club games.

its usually only local bookies that do odds on club champo games and you can be sure as hell if you have inside info they do too. the biggest value of the year is to be had in the pre - season inter county matches coming up in the next month. As far as I know PP and Boyles give odds on these games and sometimes Betfair but it can be a bit hit and miss. Apart from one slip last year where I done my rectum on a DCU / Meath game Iv beaten the bookies pretty bad over this time

[Mullicker]

The normal process with something like this would be to define the model loosely, now apply it to a full fixtures games (this would involve compiling the same data for all of the teams, stats etc)
Through trial and error youll eventually find yourself agreeing with most of the market prices. Its now youll know you've a happy medium of the variables, their strengths and confidence in your own model to start weighing in behind in bets.

Once you adjust accordingly for short term goals scored and motivational aspects then it should work for there on in.

Why go to all the trouble of coming up with a model with the goal of being inline with the market, i can see that it will be a good indicator you are in the right ballpark, but if you were serious about coming up with the best possible model and finding an edge, back testing your model against historical data is the only way to do it no?

[hotspur]

For anyone who wants to take some of the fun out of it ...

A Birth Process Model for Association Football Matches:

http://www2.imperial.ac.uk/~ejm/M3S4/Problems/football.pdf

Modelling association football scores and inefficiencies in the football betting market:

http://www.math.ku.dk/~rolf/teaching/thesis/DixonColes.pdf

A random point process model for the score in sport matches:

http://www.utia.cas.cz/files/Soutez_09/Volf/VOLF_IMAMAN2009.pdf

Modelling football match results and the efficiency of fixed-odds betting (at least goes into the factors which should be taken into account even if it choose to eschew the traditional prediction method of forecasting the winner via relative goals scored:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.65.9890&rep=rep1&type=pdf

PREDICTING AND RETROSPECTIVE ANALYSIS OF SOCCER MATCHES IN A LEAGUE:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.56.7448&rep=rep1&type=pdf

[Lurker23]

I'll be doing something similar with US Sports next year. Going to try to handicap baseball, american football and basketball games. More of a hobby than anything else.

I've about 10 books ordered at the moment for different sports. I'm going to scrape data off the boxscore pages of certain websites using Python and set up a database. I'll worry about statistical analysis when that's done.

I'm most excited about receiving this book:



It's received whopper reviews. I don't think I'd have the time to try a football model atm but it's something I'd love to attempt if I make good progress with the US sports.


Good luck with this anyway!

[TommyGunne]

I'd defo be interested.

I've a small bit of the groundwork done, I don't have much time these days, but have a good idea of how the pros model these properly, and have no problem doing it that way.

I'd be interested in doing it to get a good model going anyway, cos have intended to do it for a long time. This should hopefully give me the motivation to do it.

If I have the time to put into the model then I'll be happy to lay/based good size based on it too.

[ShipIt]

I think its far from a no brainer market to compile. I think its a market that gets punted in a lot by the general gambler who doesnt really have a clue what price it should be, or what exactly they should be looking for. Its something im going to look at myself to try and see how exploitable it is as im aware the bigger more professional punters have already identified it as a good market.

The mug punter being 20 pips out, the diff from 1.9 to 1.75 is were your going to get your margin. It leave you the recommended % margin for error and give you a valuable bet!

Im already doing some betting myself on match odds - but with the prem being so heavily liquid its very hard to find too many errors. If anyone has questions about how im doing the match odds, pm me.

Start off Im realy drunk......ill leave it it that...my name is Karl I hope to not forget thids thread.

[ShipIt]

Without being too much of a prick do u think u can beat the Pros (Bloom one example) on markets they own.
2.5 goals is The most statistical accurate market.

[ShipIt]

Hey Im a drunk c1unt dont ever back down.
Im all for initiative.
GL.

[outdraw]

If ye come up with a model that ye would like to test against some of the upcoming major european games the games below are games that my model has identified as further from the mean and so would be easier to analyse.

The games with the * are the games I am particularly interested in.

SPOILER

German Bundesliga (15th):
Overs:
Nurnberg v Borussia Monchengladbach
Bayer Leverkusen v Borussia Dortmund
Schalke v Hamburg
Eintracht Frankfurt v Hannover
Stuttgart v Mainz
St. Pauli v Freiburg

Unders:
Wolfsburg v Bayern Munich

Italian Serie A (6th):

Overs:
Udinese v Chievo*

Unders:
Brescia v Cesena

French Ligue 1 (15th):

Overs:
Paris Saint-Germain v Sochaux
Lens v St Etienne

Unders:
AJ Auxerre v Monaco
Stade Rennais v AC Arles*
Marseille v Bordeaux
Nice v Lille

Spanish Primera Division (2nd):

Overs:
Barcelona v Levante
Valencia v Espanyol
Villarreal v Almeria
Real Zaragoza v Real Sociedad

Unders:
Athletic Bilbao v Deportivo La Coruna

[Semibluff]

Without being too much of a prick do u think u can beat the Pros (Bloom one example) on markets they own.
2.5 goals is The most statistical accurate market.

you mention any mug could gamble reasonably accurately in this market. Yet you stay its the most statistically accurate market. Do you think the mugs use stats or do anything bar looking at the last 4 games (or maybe the league table (full format of course!!!) )

Good job its not Bloom who we're looking to bet against. He bets in asian markets and not Betfair.

If it turns out it doesnt work and the "model" we reach at the end of the whole process is too close to the market. Then I/we'll have lost nothing more than time!

Open to all ideas on how we could work this - will be starting it myself next week!

thanks for replies so far


PS - hope the hangover was a killer

[bustamoves]

good luck with this semibluff. sounds interesting and am interested to see how it goes for ye.

always been interested in modelling but never actually tried it. would love to try develop a model around BAGs greyhounds ever since a bit of a market/liquidity developed on betfair these last couple years.(do u think i'm mad?)

dunno why SHIPIT is so cynical. how can you compete with 'the pros' in a market if you dont try develop a method, first off with some decent model? keep going, gl, nothing to lose. and time invested will surely stand in terms of experience gained, right.

[outdraw]

-Results at home vs similar teams
-Results away vs similar teams

How do you decide what is a similar team?

[Mullicker]

How do you decide what is a similar team?

Last year i had something like:

The big three:
Chelsea, Arsenal, Man U

4th place chasers:
Liverpool, Spurs, City, Villa, Everton

etc.

I adjusted this through the season but these two groups were pretty constant.

[Semibluff]

thanks very much to those who pm'd me

alot to do - but looking forward to it!

[TommyGunne]

Without being too much of a prick do u think u can beat the Pros (Bloom one example) on markets they own.
2.5 goals is The most statistical accurate market.

No money in sports betting. Everyone is solid.

[outdraw]

When backtesting the model it could be a good idea to have a saver bet when the home team is superior to the away team. For example bet Man Utd to beat Stoke 2-0 as well as betting on the overs in such a proportion that if the game ends 2-0 that total winnings is equal to 1.5 times total outlay.

[outdraw]

How are you getting on with your analysis Semibluff?



I have added some more factors into my model (still very much a work in process).

Games tomorrow:

Unders:

Bari v Bologna
Cesena v Genoa
Fiorentina v Brescia


Overs:

Lazio v Lecce
Real Madrid v Villarreal
Napoli v Juventus

[Mullicker]

Why go to all the trouble of coming up with a model with the goal of being inline with the market, i can see that it will be a good indicator you are in the right ballpark, but if you were serious about coming up with the best possible model and finding an edge, back testing your model against historical data is the only way to do it no?

After doing some reading, coming up with a model that gets close to the closing line(of a sharp bookie) 70% of the time would mean you have a really solid model. Would be interested in exchanging ideas and building a model but i am a little busy right now.

[outdraw]

Unders:

Bari v Bologna
Cesena v Genoa
Fiorentina v Brescia


Overs:

Lazio v Lecce
Real Madrid v Villarreal
Napoli v Juventus

5/6


I'm getting closer to combining all the relevant factors into a good weight ratio but the data collection is taking a long time.

I'm working on the games on the 15/16 for the 5 major leagues.

Has anybody had any experience with a virtual assistant?

[outdraw]

Weekend games:




Unders:


Stade Rennes v Arles
Brest v Caen


Overs:


Bayern Leverkusen v Borussia Dortmund
Werder Bremen v Hoffenheim

[Semibluff]

havnt forgotten about this. Really am intrigued by it, and seeing as im self classified as being "maths dyslexic" i would have worked my own longer way around to basing some of these models and variants.

The maths used behind these complex equations is alot more than im used to, so banging heads with a few guys who are good at these models and hopefully can get something together.

I think getting the info/model in place although hard work, should be the easier bits!

Its weighing in the external factors and giving them relevant weights that will be alot tougher.

[outdraw]

Weekend games:




Unders:


Stade Rennes v Arles
Brest v Caen


Overs:


Bayern Leverkusen v Borussia Dortmund
Werder Bremen v Hoffenheim

Werder Bremen v Hoffenheim - no bet (Almedia (Bremen top scorer) has gone to Besiktas).

[outdraw]

I'd defo be interested.

I've a small bit of the groundwork done, I don't have much time these days, but have a good idea of how the pros model these properly, and have no problem doing it that way.

Can you describe how the pros do this or give a link to a website with a detailed description?

[Bubbleking]

Testing out a model tonight in the Chelsea v Bolton game. Have the unders on 2.5 goals at 1.98. If successful I may run it for other games later in the week if I have time

[The Situation]

Testing out a model tonight in the Chelsea v Bolton game. Have the unders on 2.5 goals at 1.98. If successful I may run it for other games later in the week if I have time

You shouldn't judge the efficiency of your model just based on a one game sample tonight.

[Bubbleking]

True. I think its a good rough modelbut needs a bit of tweaking over time

[Mellor]

When backtesting the model it could be a good idea to have a saver bet when the home team is superior to the away team. For example bet Man Utd to beat Stoke 2-0 as well as betting on the overs in such a proportion that if the game ends 2-0 that total winnings is equal to 1.5 times total outlay.

That's rather foolish imo. If you are punting one bet in such a proportion to cover total outlay, it seriously affects you profit when it misses.

Basically, for a long term model (as opposed to once of bets) its a bad idea to have opposing bets on the same market. Multiple bets are fine if there aren't related.