mathare's gamblog

[mathare]

31st October 2009
I don't know if it ever happens to you but I experience what I experienced late last night several times a month I'd say. I had gone to bed a bit later than normal, watched a bit of telly and read a few pages of a magazine before deciding it was time to sleep. So light off and I'm getting ready to doze off. My mind is steadily being cleared of all the things I had been thinking about during the day and then BANG! I was wide awake and I didn't really know why at first. There had been no noise outside or anything like that. My brain had been steadily processing thoughts and data for ages and out of nowhere came up with something that jolted me awake. An idea that on the face of it seems so stunning that it possessed me and prevented me from sleeping until I had thought it all through properly and worked out how best to act on said idea. It made me want to bound out of bed, fire up the PC and start testing this idea but I knew that wasn't the best thing to do. But all the same I couldn't let go of this idea. I wanted to know more, do some research and start testing it but I knew I needed sleep. My body said it was tired but my brain was firing on all cylinders and wouldn't submit for hours. Would I still remember all the details in the morning? Eventually I did get a few hours sleep but woke early with this idea still buzzing round in my head.

So what was this idea that kept me awake for so long last night? Another idea for a football betting system and one that seems so pure it's hard to see how it can fail. Now I just need to put in the effort to test this idea using some real data and see if it does indeed have legs or whether it's just another good idea that doesn't work in practice.

[MattR]

Intriguing Mat, will watch developments with interest! As a teaser, can I ask what market it is?

[mathare]

Intriguing Mat, will watch developments with interest! As a teaser, can I ask what market it is?

It could extend to several markets, for the major leagues at least where more bookies prices are available, but initially I am looking at the match result (H/D/A) markets. It's the main market for any football match and one I am keen to crack if at all possible. That said it may eventually lead to me looking at other markets such as clean sheets, win to nil, correct scores and perhaps even handicaps or the supremacy spread markets

[mathare]

5th November 2009
The football idea has been put on hold for a little while as I really haven't felt up to giving it the attention it requires. I had a quick play with it in Excel and the basic system isn't profitable for draws (which was the easiest case to look at so that's where I started) but there are many variables I can tweak that will affect the ROI so I need to draw up a proper VBA-based workbook to test the idea out fully and logically. I think there is going to be at least one profitable system in there somewhere, I just need to put the work in to find out.

I had another of my moments of clarity too the other day. It was while I was in the shower (where I have a lot of my best ideas it seems). Remember I mentioned the system analysis function I had added to my betting records spreadsheet? I plan to extend it to show me more stats including the max/min values the bank reached along with when these figures were attained plus a complete breakdown of the profits by year, month and day of the week. These breakdowns will only be meaningful for systems with lots (1000s) of selections but I want to see if there are particularly good or bad months, days of the week that are better/worse than others and also to see how the ROI, SR etc changes year on year. All the information I need is in the spreadsheet I just need to write a few lines of code to bring it all together. As you know, I love my stats and this is precisely the sort of information I want at my fingertips.

Finally, I wanted to give my own False Favs laying system a bit of a plug. It's on a hugely profitable run at present and the qualifiers are being dished out for free on this forum for all to follow. You could do worse than give them a look. I have made over 80pts profit to Betfair SP in under 500 bets. I started following them myself with real cash on 21st August this year with a £500 bank, which now stands at just under £800 using a steady, sensible staking plan (more details on request). That's a 60% increase in a little over 2 months! I use Betdaq rather than Betfair but my recorded ROI on these is 3.98% which is damn good for a laying system. I have attached a graph of the profits to both SP+10% and BSP against number of bets so you can see how things have progressed. It's a system I am very proud of and want as many people as possible to benefit from it.

[mathare]

7th November 2009
I have those yearly, monthly and daily breakdowns now and they look good. I have already spotted a few interesting things, although it must be noted that the samples are a but small for some systems to draw any really significant conclusions. That said, the following statements look reasonable based on my data:

• The Mark Johnston system does best between June and August with the performance outside this period fairly poor, with the possible exception of November actually.
• The Michael Stoute system has weakened from a strong showing in 2007 and performs best in August/September
• NH Media runs at a negative ROI on Saturdays and Mondays but shows a ROI around 30% Tuesday to Thursday
• Double Top's SR is lowest midweek (Wednesday to Thursday) with Wednesday actually showing a loss. The overall SR seems to be improving slightly year on year though
• The Lay'em system is least profitable on a Saturday. It also shows a drop in SR (and a loss in actual fact) in April and September, presumably as this is roughly where we start to see a changeover from flat turf to winter AW.
• NH Profit Machine lays show a loss on a Wednesday (I really don't understand why) and for the month of June (weaker summer jumps?)

Remember, this is all based on the data I have accumulated through my betting records and as such should not be taken as gospel. Nor am I suggesting anyone should not follow a system at certain times of the year or on certain days of the week but I provide this information for any people who have recently started these systems so they have an idea of what to expect.

[mathare]

14th November 2009
I have recently become consumed by ideas for football betting systems, rating systems and so on. My brain is always ticking away, even at night when I am trying to sleep and it is has got to the point where it has started to keep me awake so I have succumbed and started to get some of these ideas down on 'paper'. I have various Notepad files on my PC full of various notes plus numerous spreadsheets in various states of completion. I'm pretty sure I am going back over some old ground I covered a few years back but I am doing so with a definite purpose this time and a real sense of knowing where I am going with this and what I am looking for, something I may have lacked last time I started down these routes.

I'm trying to concentrate on the simple markets and using fairly basic data to accomplish my goals. I figured the only way to make a profit from betting is to have a different opinion to whoever is setting the market prices (be it another punter on the exchanges or the odds compilers at the bookies) and for my opinion to be right and theirs to be wrong more often than not. OK, that's not strictly true as once a bookie has put his line in the sand he may move it based on market forces rather than changes in his opinion. For example, a bookie may feel a match is more likely to go over 2.5 goals than under so sets his lines at evens for unders and 4/5 for overs. If he is inundated with unders bets at that price he may shift his lines to 9/10 for both sides. Now that makes it seem as though he has no idea whether the match will go under or over but the odds compiler probably still thinks it will go over - it is the weight of public opinion (backed my their cash) that has forced him to rethink his market prices but perhaps not his opinion. So I need to either swim against the tide or disagree with how the bookie sees the match playing out. I'm sure they have more stats and software at their disposal than I do but that doesn't mean we can't make a fair fight of it if I can apply my thinking in different or better ways than they can.

Maybe at the end of all this data crunching and analysis I will come up with nothing. Maybe. But I think there'll be something in all of this that will make it worth me spending the time on it. In fact I actually see a very bright future for all this work in some form or another :wink

[barrelmaniac]

A very thought provoking post regarding the bookies price, opinions and market forces. I can see exactly where you are coming from.

[mathare]

Sunday 15th November
It's not taken long but already I have lost my way with this football betting stuff. I have devised a rating system and applied it to a load of data (14 seasons worth of data from the English leagues) but now what? I thought I knew what I was doing with it but it turns out that I am no longer so sure. My brain is still rather muddled on all of this and it is taking some time to get everything in to order.

I have started to keep an A5 notebook by the bed for thoughts that occur to me as I am trying to drop off to sleep and last night I made over three pages of notes on ideas for football systems and lines to research. Whether any of these lead to anything remains to be seen. Whether they actually make much sense in the cold light of day also remains to be seen if I am honest. But none of this gets me any closer to working out what I am doing with these ratings I have put together. However, I am not going to be defeated. I have books on football betting I have read in the past. Recently I dug them out and put them near the top of my 'to read' pile so perhaps I need to get round to those sooner rather than later. I feel as though I just have a hurdle to get over and then I can gallop on with this for miles.

[mathare]

Monday 16th November
I am starting to get back into it now. Yesterday I have a brief moment of clarity that allowed me to get over one of the hurdles I mentioned previously and now I am running with a particular idea to see where it takes me. It's taking my CPU to new heights as it is involving quite a lot of number crunching so while my PC is busy doing it's bit I am reading more and more about football betting, and making more and more notes. I now have numerous pages of notes in that A5 notebook of mine and need to find a way of collating these notes into a more useful form as well as categorising and prioritising them. I think many of these notes I have made will lead to ideas that blossom and bloom into a potentially massive area of investigation so I need to be sure that I do things in the right order as some will surely unlock data that I can use in later research.

I need to find a way of focusing my mind better too. I find that when I write things down I get them clearer in my head so maybe that's what I need to do here. Perhaps I need to start documenting the full approaches I am taking with each line of investigation such that my brain can process it properly, spot any flaws in what I have done and also identify areas into which I can extend the research.

I have to admit though that I am amazed at the number of ways in which I reckon I can slice and dice the raw data I have to give a new angle on betting markets. It's surprising what can be done with a few match results and betting odds.

I will update this thread with summaries of my findings as I have them. At present I am still working on some rating system ideas and looking into ways of optimising the raw rating system as it uses several inter-related variables. It's testing my brain as well as my CPU!

[mathare]

Wednesday 18th November
OK, now I am confused. I have my head on straight when it comes to where I am going with all this football system research and I have a plan for examining all this data that will probably take me several months to work through, even if I work on it full time (which I can at present, pretty much but I may be under pressure soon to get a 'proper' job ). But this morning I tested an idea using matches from the 09/10 season so far and seem to have uncovered a system with a ROI of around 30%! But it only works for home teams even though the model I created should apply equally to home and away teams given the way it was built. So why does it only work for home teams? Is the model flawed or are we seeing a significant imbalance in home and away odds from the bookmakers? Is this the home advantage coming into effect through the odds? I must admit I don't know at this stage but I will be doing all I can to try and work it out.

The headline stats for this system though, if it holds up are 97 bets (out of 692 matches played this season in the English league) of which 37 were winners netting a profit of 30.19pts which is a ROI of 31.12%. Looks too good to be true doesn't it? The data sample used is a bit small for this to be a truly meaningful test, and I need to try to find out why the away side of the system runs at a ROI of nearly -23%, but it's an early glimmer of hope that the work I am putting in to this project may eventually pay off handsomely. If I can't find any obvious flaws in this system for home teams I will probably start another live trial here on the forum, perhaps till the end of the season, to see how things stack up. Maybe we've just been lucky with the results so far this season and things will normalise over the remainder of the season but we'll see...

[MattR]

Interesting Mat. Can you check it on another couple of leagues from the football data site, perhaps la liga and serie a (or just check the 08/09 season for the English leagues) to see whether what you are seeing pans out the same there? My thinking is that if it does come out similarly then you're more likely to have something there of substance and it's the odds that are the imbalance.

[mathare]

Interesting Mat. Can you check it on another couple of leagues from the football data site, perhaps la liga and serie a (or just check the 08/09 season for the English leagues) to see whether what you are seeing pans out the same there? My thinking is that if it does come out similarly then you're more likely to have something there of substance and it's the odds that are the imbalance.

There is nothing to suggest the model would apply exactly as it is to any foreign leagues but what I can do is, as you suggested, move some data out of that used to build the model and into the data set used to test it, which is what I am looking at now. The original model was built using all data from the 95/96 to 08/09 seasons and then tested on the 09/10 data but I am investigating now models built on the 95/96 to 99/00 data and tested on 00/01 to 08/09, models built on 95/96 to 00/01 and tested on 01/02 to 08/09 etc. I obviously can't test the model with data that was used to build it in the first place as that won't give me an independent test. It takes a bit of computing time to rebuild the models and then test them though. I might have some answers later on this afternoon though.

Each test I have done so far picks out many more away bets than home bets though and I have yet to really fathom out why. Suppose there is some inbalance in the odds - why is it there? Why am I finding so many more away teams at value odds? Are the chances of an away win being underestimated by the odds?

[MattR]

So if I'm getting this right Mat, it is selecting a lot more value away bets but the ROI is significanty worse and indeed negative. Hmm, without knowing how you are coming to a decision of value I'm not sure. Is the model you are creating analysing just home data to get value home odds and vice versa for away, or is all the data being used to arrive at a value odds decision? It could just simply be the fact that there are far fewer away wins than home wins so, value or not, it's harder to pick away wins than it is home just by the sheer numbers alone. % wise how are the strike ratesof winnerson the system regarding homes v aways?

[MattR]

Each test I have done so far picks out many more away bets than home bets though and I have yet to really fathom out why. Suppose there is some inbalance in the odds - why is it there? Why am I finding so many more away teams at value odds? Are the chances of an away win being underestimated by the odds?

Just thinking on this Mat.

The fact that it's picking so many away bets would lend itsef to being very hard to show a profit I would have thought due to the significantly fewer away wins there are in a season.

[mathare]

So if I'm getting this right Mat, it is selecting a lot more value away bets but the ROI is significanty worse and indeed negative.

Right.

Hmm, without knowing how you are coming to a decision of value I'm not sure. Is the model you are creating analysing just home data to get value home odds and vice versa for away, or is all the data being used to arrive at a value odds decision?

Each team in every match is given a rating. These ratings are then split into a number of bins. I then count how many teams with a rating in that bin have won, drawn and lost. Repeat for all bins. Graph the data, add a best fit line and you have a chance of the team winning, drawing or losing for any given rating. Use that information to work out value odds for each team in the test data winning their match. If the available odds exceed the value odds then we have a bet. So all the data, home and away, is being used to form the value decision, in effect. I have thought about splitting it home and away though, but that's more effort at this stage. One for the 'to do' list.

It could just simply be the fact that there are far fewer away wins than home wins so, value or not, it's harder to pick away wins than it is home just by the sheer numbers alone. % wise how are the strike ratesof winnerson the system regarding homes v aways?

You could be right. In terms of SRs they are fairly different. The home SR on the original test (over the 09/10 data) was 38.14% with the away SR coming out at 21.37%

[mathare]

The fact that it's picking so many away bets would lend itsef to being very hard to show a profit I would have thought due to the significantly fewer away wins there are in a season.

Perhaps, yes. But it's why it is picking so many away bets that is puzzling me most. Why would there be 4 or 5 times as many value bets for teams playing away than at home? That suggests to me there must be some factor in the odds I am overlooking.

It's worth noting that in some of the other tests I have done the away ROI isn't always worse than the home ROI although the ratio of home to away bets remains around the same.

[MattR]

. I have thought about splitting it home and away though, but that's more effort at this stage. One for the 'to do' list.

I think that would be interesting to see the difference it would make but like you say, one for the to do list given the extra work involved setting it up. Personally I seperate home and away records for my systems but that's just me, not saying that is right or wrong. The one's I use from the goals for/aga probabilities for example uses the home teams home for/against and the away teams away for/against.

[mathare]

18th November 2009
I have put the hours in on this one today and concluded that what I saw earlier was indeed a fluke :(

Let me briefly describe what I have done to show this as someone may spot a flaw in it, and it will help me mentally arrange my work and understand it better.

I have data from 14 seasons (95/96 to 08/09), plus whatever has been played so far of the 09/10 season. I have an idea for a model built around a ratings system. I want to test this model. Obviously I can't test it with data that was also used to build the model as I am then testing the model against itself, in part at least. So I need to split the 14 seasons of data into some for building the model and some for testing. I could randomise which seasons are used for which part of this but I chose instead to keep the systems in order and use the first X seasons as inputs into the model and the remainder as test data.

As I said earlier the model revolves around a rating system. Basically I rate every team in every match, divide these ratings into a number of bins and then for each bin I look at how often a team with a rating in that bin wins their match. This data is graphed and best fit lines added. The equation of these best fit lines can then be used to estimate the chance of a team with a rating of Y winning their match. From this one calculates value odds for that team winning which are compared to the bookies' odds. When the latter exceed the former we have a value bet. That's the theory anyway.

I binned all the data and calculated the best fits for varying numbers of seasons as inputs, working all the way from 5 seasons of input data (with 9 seasons of test data) to 14 seasons of input data (with 09/10 as the test data). It took a while but I hope it was worth it, if only to disprove something. As more seasons are used to build the model the overall ROI improves steadily, suggesting the model is getting more accurate. It remains negative throughout though so we never see a profit from this hypothesis. In general the away ROI is much better than the home ROI, which makes sense again as we see many many more away bets than home bets indicating the chance of away wins is underestimated as bookies (and punters) overestimate the chance of a match ending in a home win so we have longer away odds than is 'fair'. All good so far. There are fluctuations in the data as we add more seasons but this is to be expected. The general trends are there though. Until we use all 14 seasons of data as inputs to the model and test on this season's data.

As I mentioned in a previous entry to this gamblog that sees a strong profit from the home sides and a big loss in the aways. It completely reverses the previous trends I have seen which can mean one of two things: either something fundamental has happened in the game of football to suddenly flip the model on its head or the relatively small data sample is atypical and we can expect a correction in the future. I think it's obvious which of those two is more likely isn't it?

So while the model using all 14 seasons gives excellent fits to the real data that doesn't translate into a profitable system, at least not like this. I am still looking into other ways of using this data though.

But are there ways in which the model could be improved to make it profitable? I have examined the profits (losses) from the version of the model that used 10 seasons as inputs and 4 seasons as a test to see. This choice of model was fairly arbitrary but I thought the 10/4 split was quite a reasonable one to choose to study. I broke the profits down by division, by month and by ratings bracket. And do you know what I saw? Nothing of interest, really. The League Two away games are marginally profitable but less than 14pts from 4 seasons worth of betting - no thanks. And why should this model work better for these games than others? There is no real reason why so again I think we have a statistical fluke effect shining through. September is a very profitable month for away wins while November and January are great for home wins. So what? Statistical flukes again - there is no rational explanation for these months outperforming others. Similarly the extreme ratings show a slight profit but we are dealing with such small samples that there is nothing significant to be concluded from this.

The system doesn't work as a value system from the ratings, despite what this morning's testing seemed to hint at. At least I know now

Now I can set about ripping it up and putting it back together in other ways that might work!

[MattR]

Mat, how does the data pan out regarding the % chance you are getting for a win, rather than just when it is value, ie do those games suggesting say a 60% or higher chance of a win perform better than under 60%?

I do think that splitting it home and away, so the teams have a home rating and an away rating would be more meaningful (although who knows the end results may be just as fruitless regarding a system!) However there are so many teams who perform better at home than away this has to have a detrimental effect on the data you are using I would have thought. Fulham in recent years would be a prime example

Last 3 seasons

Home
P 57
W 25
D 15
L 19
F 68
A 65
Pts 90

Away
P 57
W 7
D 23
L 27
F 47
A 89
Pts 44


So surely their awful away form is having a detrimental effect on their rating when they are playing at home? Conversely they may also be getting rated better away based on their home record. Or am I misunderstanding how your ratings are being calculated?

[mathare]

Mat, how does the data pan out regarding the % chance you are getting for a win, rather than just when it is value, ie do those games suggesting say a 60% or higher chance of a win perform better than under 60%?

By relating the ratings to the win % through the best fits I have developed an equation that determines the win % from the rating so I only need to know one of these pieces of information to find the other. So basically what you're asking here is do the higher ratings perform better? If we remove the idea of value, as you suggest though, how do we determine the performance of a rating? Higher ratings have a higher chance of winning, sure. But without looking at odds and value how could I check performance of a set of ratings?

I do think that splitting it home and away, so the teams have a home rating and an away rating would be more meaningful (although who knows the end results may be just as fruitless regarding a system!)

I hadn't really considered the idea of teams having two entirely separate ratings - one for home games and one for away. That's an interesting idea....

However there are so many teams who perform better at home than away this has to have a detrimental effect on the data you are using I would have thought. Fulham in recent years would be a prime example. So surely their awful away form is having a detrimental effect on their rating when they are playing at home? Conversely they may also be getting rated better away based on their home record. Or am I misunderstanding how your ratings are being calculated?

My ratings are based on the age-old Rateform ratings system as this method inherently takes into account the relative strength of a side as well as rewarding away sides more for better results. The home side contributes a chunk of their rating to the pot, as do the away side but to a lesser degree. Winner takes the pot which is equally split for a draw. This means an away draw is more beneficial to your rating than a home draw while an away win is much more valuable than a home win. So for Fulham a good set of home results will increase their rating at a slower rate than good away results would have done while poor away form will reduce the rating slower than poor home form. I will consider completely separate home and away ratings though so see what effect that has on things.

Thanks for chipping in with even more work for me

[MattR]

If we remove the idea of value, as you suggest though, how do we determine the performance of a rating? Higher ratings have a higher chance of winning, sure. But without looking at odds and value how could I check performance of a set of ratings?

Well what I meant was if for example, using your equation to find your win % , would you make profit if say all those that had say a 60% win % or above were the bets for the day? Regardless of whether your win % was above or below what the actual days odds were?

Mat, something just came to me then, if this is correct - this win% you are getting, say it's 75% is that then converting to 1.33 (1/3) odds? Well, presumably that means the other team had a 25% win %. I think maybe I've just found the 'flaw' as you put it. There is no accounting for a draw in your 'market' so effectively aren't you pulling odds for a 'draw no bet' market? This would account for why so many away games are value.

My ratings are based on the age-old Rateform ratings system as this method inherently takes into account the relative strength of a side as well as rewarding away sides more for better results. The home side contributes a chunk of their rating to the pot, as do the away side but to a lesser degree. Winner takes the pot which is equally split for a draw. This means an away draw is more beneficial to your rating than a home draw while an away win is much more valuable than a home win.

Ah ok, yes I remember those.

Thanks for chipping in with even more work for me

You know you love it

[mathare]

Well what I meant was if for example, using your equation to find your win % , would you make profit if say all those that had say a 60% win % or above were the bets for the day? Regardless of whether your win % was above or below what the actual days odds were?

OK, I get you. I'll have a look at that this afternoon if I get a chance.

Mat, something just came to me then, if this is correct - this win% you are getting, say it's 75% is that then converting to 1.33 (1/3) odds? Well, presumably that means the other team had a 25% win %. I think maybe I've just found the 'flaw' as you put it. There is no accounting for a draw in your 'market' so effectively aren't you pulling odds for a 'draw no bet' market? This would account for why so many away games are value.

I'm not sure that's the flaw, if there is one. All I do is work out the chance of a team winning. So if that's 75% as you say then there is a 25% chance they won't win. They could draw or lose - just not win. The draw is factored in. In fact I have built models for winning, drawing and losing but I chose only to value test the winning model as the draw model is very much weaker and the losing model is much harder to test as you have to effectively dutch the draw and opposition winning.

[MattR]

Ok I'm with you now, I thought in that example that you just had a 75% chance of one team winning and a 25% of the other team winning, not meaning that it would be 25% of a draw or loss.

[mathare]

19th November 2009
I have looked into that idea MattR suggested earlier - backing teams with high ratings regardless of whether they are value bets or not. I decided to start with teams rated 1500 or above as I had to start somewhere. Teams rated at 1500 have a 44.23% chance of winning according to the model I used with teams rated at 3000 have a 55.50% chance of winning. I looked at home teams and away teams separately, placing 1pt level stakes bets on all teams with a rating greater than or equal to 1500, 1600, 1700 etc. working in intervals of 100 up to 2000 and then 200s up to 3000.

The attached graph shows the ROI for home teams, away teams and total plotted against rating. The most profitable rating range is 2500 and above but there are only 472 bets in this range generating a profit of 11.81pts which over 4 seasons is nothing worth writing home about so I think this idea is basically dead in the water. I'm fine with that though. I expect to hit several dead ends in this project but my attitude is that if I don't investigate these avenues then I will never know whether or not they could be my route to riches.

[mathare]

20th November 2009
I had yet another idea on this last night - what about other staking plans? In particular the fixed profit staking plan. With level stakes I am risking the same amount on a longshot as I am on the shorter priced selections so what if I vary the stake with the odds and look to win a fixed amount with each winner?

I looked into this idea this morning and it's a significant improvement, increasing the ROI from the model by some 4 or 5% but it still doesn't make it profitable. No staking plan can turn a losing system into a winner but it is worth knowing that fixed profits could be something worth considering for other systems I come up with on this epic journey through umpteen season's worth of data and systems.

[MattR]

Interesting Mat, I remember when I first found this site and used the staking system Keith had on his main w2w page (not sure if it's still there) It was a stake linked to the odds, not quite a fixed profit, but basically the lower the odds the more you bet. That worked quite well on a couple of horse systems but with the lower football odds I'm not sure how it would fare.

Just had another thought to throw your way, not that I want to give you more work :D Have you had a look at all at how things fare if you take the biggest difference between the two team's ratings? For example those with say a 1000 gap or more in the ratings , home team 2450 away team 1200 etc etc

[mathare]

Just had another thought to throw your way, not that I want to give you more work :D Have you had a look at all at how things fare if you take the biggest difference between the two team's ratings? For example those with say a 1000 gap or more in the ratings , home team 2450 away team 1200 etc etc

Shhh, you're spoiling the next chapter

You're right. This system doesn't take into account the strength of the opposition, only how highly rated the chosen team are. The odds will take into account the strength (and form) of the opposition to some degree but that could just mean we are picking out value bets where there isn't actually that much (if any) value as the opposition are too strong for the chosen team. Factoring in the relative strengths of the teams is (probably) the next step in my football odyssey.

If I am honest, I didn't expect this system to work (and it doesn't) simply because it focuses on the wrong things - one team only and not the quality of the team they face. But that won't stop me pulling it apart in every way I can think of. I still have a few more sub-ideas to work on for this team rating system and then I will give up, but not till I have really done it to death to make sure there is nothing in there. I plan to leave no stone unturned.

[MattR]

Look forward to seeing what you discover next. If there's something in this I think you'll track it down! Fascinating thread as always Mat

[mathare]

20th November 2009 (part two)
You have probably guessed by now that I don't give up easily on an idea, not until I have looked at it from every angle I can think of and done all I can to see if there is any chance at all of it producing a profitable betting system. And so it is with this team rating idea.

The system I am trying to build has the concept of value betting at it's heart. A value bet exists when the available odds exceed what the model estimates as the fair odds for an event. But the odds released by the bookies may include many factors that are not built into the model producing the fair odds, such as team news, injuries etc. Suppose the odds on a team are much higher than expected for a reason, a reason unknown to the model. In such a case we have underestimated the true odds of that team winning and thus overestimated the value margin.

What I have done is breakdown the profits from one of the models by value margin, looking at bets above a certain margin. So for a 10% value margin the bookies odds must be at least 10% greater than the fair odds and so on for other margins. And? Nothing significant really. At the extreme margins (100% and greater ie the available odds are double the 'fair' odds) the ROI drops quite a lot but there are also few bets at this level so the sample is small which will have an effect on things. The overall ROI is best at the smaller value margins of 5-20% (which suggests that the big prices are big for reasons that are indeed not factored in to the model) but only at 10% is the ROI better than the basic system.

I have also turned this on its head and looked at the profits and ROIs when the odds are equal to or less than the same value margins as used above. The reasoning behind this is that if there are factors we haven't accounted for when calculating the true odds then we should be able to identify a value margin above which we shouldn't bet. Can we do that? Not really. By cutting out the bets where the value margin is huge we do improve profits but not enough to turn this system into something worth paying much attention to. The bets figures are for away bets with a value margin of 40% or less (ROI of 2.14%) but there is no significant reason why this margin would work and others around it wouldn't so I don't think this is a genuinely profitable system long term even with this value margin filter.

[mathare]

20th November 2009 (part three)
How I could be so stupid? :splapme

I looked at teams whose margin of value exceeded certain figures and also teams whose value margin was less than or equal to a certain figure. What did I miss? The inbetweeners! Those teams whose value margins lies in a certain range - above one figure but less than or equal to another. A quick bit of spreadsheet trickery to test this on the most likely range and we have it - a 7.90% ROI with 1514 bets over 4 seasons. The system is worth around 30pts of profit a season - and that's worth having.

But before we jump up and down and start celebrating we need to make a couple more checks:
i) how does it look for this season so far?
ii) will it stand up in the real world?

The first check is an obvious one but what does that second one mean? This is all based around value, right? The odds I have used to determine the margin of value are what I have dubbed ave-max odds - the average of the average and maximum available odds. I've used these as I have shown before that it is usually possible to beat the average odds by shopping around but not always possible to achieve the maximum odds as they are often with some obscure foreign bookie. So were I to follow this system for real would I be able to get the same results with the prices I can find as they are integral to whether or not we place a bet? I don't know the answer to this for sure and the only way to find out is to paper trade it, which I will do IF it looks like it holds up OK for the 09/10 season so far...