System Advice

[AndrewKent]

Hi there, I've had a lot of cracking advice over the last few days from people on here regarding betting systems. A lot of the advice has been sound and what I have learned is that there isn't any simple or sure fire way to design a system and many factors have to be taken into consideration to make the system as accurate as possible. For example, someone on here said that there is a MASSIVE difference between "Team" data. I was looking at names on a spreadsheet, and not taking into consideration the "Team" in the sense of who was playing. (For example, the Man Utd "Team" for the Champions League final would be different in many ways to the Man Utd team playing Oxford in the League Cup). So with that in mind I want to try and teach myself a little about factors for systems. I have the fantastic data source from www.football-data.co.uk and using that I want to start small by analysing previous results. Here is what I was thinking. Kilmarnock vrs Hearts Last 6 games Kilmarnock results in the last 6 games Hearts results in the last 6 games Kilmarnock results in the last 6 home games Hearts results in the last 6 away games Current Season Kilmarnock results this season Hearts results this season Kilmarnock results in this seasons home games Hearts results in this seasons away games Last 5 years Kilmarnock results Hearts results Kilmarnock results in home games Hearts results in away games Head to Head results between Kilmarnock vrs Hearts in the last 6 meetings results between Kilmarnock vrs Hearts in the last 5 years Now I know this is all basic calculations and might not turn up anything massively helpful but I want to get a feel for doing this. Can anyone suggest changes to what I have listed? Can you list these in order of importance? Any amendments in general. Like I said, I am only learning and still finding my feet so any form of help would be very much appreciated. Kind regards, Andy

[Mulkis]

Re: System Advice Andrew, I've been reading your recent treads and the questions you are asking they all are valid and sound. But I've got a feeling that you somehow missing to raise and ask one main question for yourself: what exactly are you trying to dig from all this data?. Are you trying to find probabilities (chanses) of certain event to happen (Team A win %) or are you trying to identify certain patterns indicating that some outcome might be a "winner"? I think this is important to answer these questions in the beginning otherwise it is possible to manipulate all this data in a million of different ways with no substantial results.

[AndrewKent]

Re: System Advice Mulkis, you ask a very valid question. I have had a limited amount of success "laying" results based on the fixed odds of betting companies. I say limited because although my picks have been 90% accurate, the odds are usually quite low so it does mean when one bet does lose, it wipes out the majority of my profit. I have been using this system for a couple of months and it only has about a 1.5% ROI. I want to analyse other aspects of previous match outcomes to see if I can improve my chances of "laying" a bet. I want to see if I can find a pattern which will allow me to bet on games which have better odds so I'm not always Laying the clear under-dog. Does this make sense? Andy

[Mulkis]

Re: System Advice I'am not sure I've understood your concept, but regarding what makes sense I think it is generally two concepts: first when you try your best to find "true odds" of the event to happen and then try to bet on better odds for 2-3% ROI or, second, to indentify certain more or less stable patterns indicating to winning positions with expectation of 10% or more ROI.

[AndrewKent]

Re: System Advice Okay folks, I've finished the basic analysis and I have calculations to look at the following measures: Kilmarnock vrs Hearts 1) Head to Head in the last 10 years 2) Head to Head in last 6 meetings 3) Hearts results in the last 6 away matches 4) Hearts results in away matches in the last 10 years 5) Kilmarnock results in the last 6 home matches 6) Kilmarnock results in home matches in the last 10 years 7) Hearts results in the last 6 matches 8) Hearts results in the last 10 years 9) Kilmarnock results in the last 6 matches 10) Kilmarnock results in the last 10 years Although this is all basic checks, I'm looking to learn how to do this and I wonder if someone could give me their opinion on these measures. Should they be adjusted? Should I discard some, keep others? Are there other basic measures from previous match results that are more important? Cheers, Andy

[Datapunter]

Re: System Advice

......... in the last 10 years

By what reasoning do you think that results from 10 years ago have any validity in any way or form in any kind of prediction model ? (in the case of football teams)

[jtw1]

Re: System Advice Purely a personal point of view , but i'd be tempted to drop the 10 year stats EXCEPT the h2h [ important imo ] I can't see what looking back over that period will achieve , esp as we have the situation up here where due to the dominance of Rangers and Celtic most teams will have a major non-win strike-rate against those 2 . Recent 6 match form , h2h's , home /away form for this season [ignoring the Old Firm games which will sort themselves out no matter how poor a spell they go through ] are the basics i would look at . GL :ok

[Lardonio]

Re: System Advice

Okay folks, I've finished the basic analysis and I have calculations to look at the following measures: Kilmarnock vrs Hearts 1) Head to Head in the last 10 years 2) Head to Head in last 6 meetings 3) Hearts results in the last 6 away matches 4) Hearts results in away matches in the last 10 years 5) Kilmarnock results in the last 6 home matches 6) Kilmarnock results in home matches in the last 10 years 7) Hearts results in the last 6 matches 8) Hearts results in the last 10 years 9) Kilmarnock results in the last 6 matches 10) Kilmarnock results in the last 10 years

You seem to have a lot of duplicate data for example with the H2H's where you both have last 6 meetings and last 10 years. That means those last 6 meetings will count twice, and they are not even necessarily relevant at all. To be honest I think your setup is far to simple for it to work - a simple form guide that will not give you any edge versus the bookmaker. Also, I would definitely scratch #1, #4, #6, #8 and #10 as what happened several years ago is not really interesting when looking at specific teams. I think you are starting work that could be fruitful but you would have to dig a lot deeper than just looking at results in the last 6 matches (I won't mention the 10 years again...) to be profitable.

[AndrewKent]

Re: System Advice Hi guys, This is exactly the kind of support I was looking for. So for that reason I am glad you are discussing this with me. As for the last ten years data, I guess the reason for using this is I thought to include everything the data source has to offer and work back from there. If the data in the last ten years isn't relevant, that actually makes life easier. Firstly it will cut the file size down massively. Is there however another data range that would be better for these initial measures? For example, results since the start of the season? Okay, based on this very basic start, I'm thinking the H2H data that is being complied is worth keeping. What about general results in the last 6 games regardless of opposition? Is that worth keeping? And moving forward, what would you suggest should be my next area for analysis? If simply looking at previous results isn't going to tell me the whole picture, where should I turn next? I have seen strategies for analysing data based on the number of goals scored. Should this be what I should look at next? Also, how relevant would analysis into the fixed odds be here? Cheers, Andy

[Datapunter]

Re: System Advice Usually when doing "research" you define a target, a goal, then you formulate a theory, often defined as a question, then you devise a test to confirm or reject the theory, then you collect whatever data needed for that test. Hmmm, what exactly is your question here ? Is it: "i have a bunch of data, what can i do with it ? " My answer: fcuk all. It's far too generic, interlaced, misleading. Number of goals, fantastic. Think about it... will you consider every goal the same, goals, own-goals, penalties. Imagine a match where the home team has 70% posession and 3 times as many shots on goal. But the away team gets lucky and manages a last minute goal against every stat of the match, 0-1. Will you consider that goal to have the same value as when the match ends 1-0 confirming all other stats. In short, you really need to move away from whatever data you already have and start putting your thinking cap on.

[AndrewKent]

Re: System Advice Seems a fair enough observation, and it also means that sadly the data source I have and have worked on is now useless. Now, based on what you are sayingni will need to source another data feed that contains much more intimate picture of each match? Such as possession and shots on/off target? Question is, where can I find a data feed that detailed?

[Lardonio]

Re: System Advice I don't necessarily agree with Datapunter here. It's true that in one match the result can often be misleading but if you have huge datasources then you will get more reliable data. I do agree that you should set yourself a target and define a goal. Also, since you are using the data from Football-data I know that you have access to statistics on shots fired, shots on target, corners etc. These numbers can tell you something about whether a result was fair or not. Considering switching your 'last 6 results' with 'last 6 shots on target' average and you are already doing something different than most gamblers. Then you can analyse odds-data and see what the average shots on target should be for a big home-favourite, evenly matched teams etc. and you can get an idea about performance. Of course, it's still only numbers and you will only get the full picture by watching the games, but the whole idea and challenge of data-driven systems is to beat the bookie by pure data analysis.

[Datapunter]

Re: System Advice

it also means that sadly the data source I have and have worked on is now useless. Now, based on what you are sayingni will need to source another data feed that contains much more intimate picture of each match? Such as possession and shots on/off target?

Not nessesarily, you need to turn it around. First you need to determine what data you need/want and ! whatfor. Or in other words what you want to do with it. Then you go and find out where you can get it.

Question is, where can I find a data feed that detailed?

As a short answer consider this: any data generally available on public websites can be collected and brought into a spreadsheet or database. You may need to pay a programmer a small fee to write a custom data collection program but you are usually talking in terms of 20-50 pounds. And as you have some programming skills yourself you might get away with a tenner as all you need is the basic starting point which you can expand yourself.

I don't necessarily agree with Datapunter here.

Sometimes even I don't always agree with me :loon

These numbers can tell you something about whether a result was fair or not. Considering switching your 'last 6 results' with 'last 6 shots on target' average and you are already doing something different than most gamblers. Then you can analyse odds-data and see what the average shots on target should be for a big home-favourite, evenly matched teams etc. and you can get an idea about performance.

Exactly. Let me illustrate with an example: Team A vs Team B You look at recent results and notice following result for their respecive last home and away matched: Team A vs Team W 2-0 win Team X vs Team A draw Team B vs Team Y 2-0 win Team Z vs Team B draw Both teams won at home with 2-0 and had a draw away. So on the face of it they are pretty equal in terms of recent result. Now add the opening odds for those matches: (fictional numbers) Team A vs Team W 2-0 win , 1.25 - 3.25 - 7.0 Team X vs Team A draw , 2.00 - 3.4 - 2.34 Team B vs Team Y 2-0 win , 4.25 - 3.10 - 1.40 Team Z vs Team B draw , 1.30 - 3.10 - 5.00 Team A had a win and draw entirely as expected, nothing special. Team B on the other hand has a result against the expectation in both matches. You can interpret this in multiple ways but you can no longer say the teams are about equal. Given the context, in this case provided by the opening odds, you cannot say both 2-0 results have equal meaning. My point is you need to put data inside a context to be able to really determine any kind of validity. Goals alone are meaningless, put inside a context they can become one of a number of valid measures.

[Datapunter]

Re: System Advice As a suggestion why don't you make an extensive list of all possible data available. Whether you can get your hand on it or not is a bridge that can be crossed later. Then come back here, post the list and ask the question: which pieces of data could be usefull and in what way ? You are likely to get 100 answers and many will actually be conflicting, o what fun. But it will provide you with a number of insights from which you make your own mind up.

[Lardonio]

Re: System Advice Love the suggestion, sounds like an interesting exercise for everyone, would love to contribute and see all the answers.

[AndrewKent]

Re: System Advice Hi guys, I have to say I am really enjoying this discussion and I appreciate your input and support. I'm learning a fair amount as I'm going along here so it's really good to have people who have already made this journey around to bounce ideas off of. Okay, in terms of next steps. I believe I do need to set myself goals. Firstly, what do I want to do? Well: "Build a program that will make LAY predictions on Match Outcomes based on statistical data only" Hopefully that isn't too wishful thinking. I know that there are many sub questions that will follow this, for example, to what level of accuracy do I want to take this, but it at least is my overall goal. As I do have a fair amount of programming skill, I think I can combine/build a data source from what is out there on the web. My questions would be more around where to find the data in a particular format (for example as an "xls" file from FootballData or perhaps an XML feed). I will go back and dig deeper into the FootballData files and see what else is in there that I can use. I'll then post up on here what I have available and perhaps someone can say what type of analysis sits best with each data type and more importantly, how each piece of analysis should be used in conjunction with each other? Can anyone suggest anything else I should be considering at this stage? Cheers, Andy

[Mulkis]

Re: System Advice "Statistic for Dummies". Have a look at it. No offence.

[AndrewKent]

Re: System Advice

Team A vs Team W 2-0 win , 1.25 - 3.25 - 7.0 Team X vs Team A draw , 2.00 - 3.4 - 2.34 Team B vs Team Y 2-0 win , 4.25 - 3.10 - 1.40 Team Z vs Team B draw , 1.30 - 3.10 - 5.00

A very stupid and simple question about this. I TOTALLY understand the point you are making about the results alone not telling the whole story. If however I was to put some sort of points system in here (lets say the standard 3 for a win and 1 for a draw) on the face of it both have 4pts. BUT if I then multiplied it by the odds of the result... Team A vs Team W 2-0 win , 1.25 - 3.25 - 7.0 = 3.75pts Team X vs Team A draw , 2.00 - 3.4 - 2.34 = 3.40pts Team B vs Team Y 2-0 win , 4.25 - 3.10 - 1.40 = 12.75pts Team Z vs Team B draw , 1.30 - 3.10 - 5.00 = 3.10pts Team A = 7.15 Team B = 15.85 ...from your example I can see doing this, this gives me a MUCH better idea who is on the greater FORM? Team B, as they seem to be playing above themselves at in the last two games. I take it the next point would be to add another layer to analyse the games themselves, so for example, the possession, and shots on target (SOT). if in Team A's 2-0 win they only had 4 SOT and 45% of possession (lets say they scored an early goal and then a goal in the 90th when Team W were pushing forward and they hit them on the break), this would make Team A's 2-0 carry even less weight than if they dominated play and had 20 SOT? Am I getting somewhere? Andy

[Martial]

Re: System Advice It sounds all like the Elo-Rating-System. But it's not profitable at all. Perhaps we should all improve this Elo-Rating-System ?!

[Datapunter]

Re: System Advice Exactly Andrew, mug punters place bets on the basis of human impressions, " they looked really good last time and i can't see them not score here " The problem is that the human brain is extremely unreliable, first we tend to remember what we would like to remember, we remember that which suits us best, then we cannot stop ourselves giving an interpretation that is entirely personal. In fact, a little phylosofical but hey ho, there is no such thing as reality, the only world us humans can operate in is the world as we interprete it. As such we all live in our own personal interpretation of reality. I'm not saying "intuition" should be eliminated completely, even where it is possible to do so. But it should be placed on a fundation of objectivity. That can be achieved by measuring the world in an objective, repeatable way. In other words you replace your human interpretation with numerical values. " they looked good last time as the time before " then becomes " they had 2x as many shots on goal, 1.8 times the time before that " Key point here is that if you look at the same data a year later the first interpretation " looked good " will probably have a slightly different meaning. After all you are a year older and new impressions of the world have shaped and changed your view. The numbers 2x and 1.8 times will always be the same. That is an objective fundation. A basis on which comparisons become possible independend from whatever braincells happen to be up for generating an interpretation. Should you base decisions solely on the basis of numbers ? That is a choice you make. At least by objectively measuring the world you give yourself an objective, repeatable startingpoint. You cannot learn where to go if you don't know where you're coming from. So multiplying the goals with the odds is maybe a crude method but it is a good illustration of the point. You need some method of measuring data inside a context. I can't give specifics as different people will apply many different method. You need to find your own way and be prepared for lots of trial and error.

[AndrewKent]

Re: System Advice This is the reason why I want to build a system. I can be a bit pessimistic about my "hunches" on a match outcome and I don't want to have to go through that "will I/won't I" thought process of putting a bet on blind. In fact I would rather not do it at all if that's going to be the case. I do however like having a statistical point of view to refer to, as it means that I no longer need to worry about my bets. As long as I have done my homework correctly then in the long run I stand to do okay. With that in mind I am looking forward to trial and error checking. I have a few ideas and questions to ask before I start. I thought of doing the following: From the "points" system we discussed above: 1) Home Team: Last 6 Home Matches 2) Home Team: Last 6 Matches 3) Away Team: Last 6 Away Matches 4) Away Team: Last 6 Matches 5) H2H: Last 6 meetings (where if the game is Team A vrs Team B, we analyse that, NOT Team B vrse Team A) Are there any checks from this that you would dismiss as pointless or are they worth trying. Also, the points that I applied were 2pts for a win, 1pt for a draw and 0 pts for a loss. Would you suggest a more detailed points system, such as, 2pts for a home win but 2.5pts for an away win? Again, I am happy to put in the work and do all of this myself but I am enjoying the conversation as it's giving me insights into building a system that I didn't even know to consider! Cheers, Andy

[AndrewKent]

Re: System Advice Okay, I've done a bit of testing today and I seem to have made a good start but I am looking to see where I can take my system next. I put one bet on yesterday as a test using this system and it seemed to work, although when I back checked I found that I was only about 70% accurate. Walsall v Notts County 1 = 2.90 X = 3.40 2 = 2.38 1) Based on the points system I mentioned in previous posts, I analyse the last 6 home games of Walsall and the Last 6 away games of Notts County, attributing points which are multiplied by the fixed odds. 2) From this I calculate the difference in points. In this example Walsall scored 11.60 and Notts County scored 13.52, giving me a difference of -2.00, a very slight advantage to the away team. 3) I then analyse all other home games where Walsall have had the rating of -2.00 and all away games where Notts County have had the rating of -2.00. The interesting thing I found was this: Home Wins = 2 Draws = 5 Away Wins = 5 4) From this I can see there is an 83% success rate if I LAY the home team. So I did, and the score finished thus: Walsall 0 v 3 Notts County So, from this, can anyone give me any pointers? Was this luck or am I on the right tracks here? Andy

[AndrewKent]

Re: System Advice PS - to keep things simple I applied a ROUND formula to the rating.

[FleetFanatic]

Re: System Advice It's one outcome of one match, You need 100s, if not 1000s to get an answer of accuracy.

[AndrewKent]

Re: System Advice

although when I back checked I found that I was only about 70% accurate.

Would checking 72,000 games be a good start you think? I appreciate what you're saying. The game in question was really just to highlight what I am do to see if anyone can point out any glaring errors from their experience. You know, something that perhaps I have missed in my calculations or something I should consider adding at this stage. Retrospective analysis is a must, hence why I started carrying this out before I even considered making the bet. What would you therefore suggest in terms of adjusting the calculations?

[kriss]

Re: System Advice I could understand researching the probability of something occuring that can be applied across an entire sport rather than a specific team or player. ie. the likelihood of there being a red card in 90s, or extra time. Or the odds on their being a 147 in a frame, match or tournament of snooker. I never concern myself too greatly with things like head to heads etc because there are so many other factors that influence the result in sport. If you did that then you'd never bet against Ronnie O Sullivan but in reality he does lose a fair amount of matches and tournaments these days.

[AndrewKent]

[jtw1]

Re: System Advice

It's one outcome of one match, You need 100s, if not 1000s to get an answer of accuracy.

It may not have been the answer you were looking for but i'm afraid's Fleet's spot on AK . Unless you can back-test , the only way is to post your selections until the end of the season and then have a look back during the summer and decide if there's potential . Keep an eye on Laying , Total goals , draws etc as sometimes these markets can help but it's only through building up a significant number of matches that will help.

[AndrewKent]

Re: System Advice I totally agree with both of you, again, I only posted that game up so show what I was doing, for someone to got "Hey, you've made a good start" or "No, change X". I agree that one game tells me nyaff all about how successful a system is.

[jtw1]

Re: System Advice Any selections for today then AK ?