Mat, pick your brain If I may?

[MattR]

Or anybody really, but I know Mat likes to get his teeth into something and analyse things.

I've been using the same basis for my ht stuff to get correct scores using poisson and have been testing things out by backing every score that is value with a £2 bet over the past month using mostly in play games from the major european leagues and brazil and argentina too.

Anyway over all in this time there have been 245 bets. The stakes per game range from around £16 (8 value bets) to a high of £26 (13) but on average it tends to be about 10 value bets for an outlay of £20 per game. To date my profit stands at £369.57 which I'm really pleased with, however since starting on the 23rd November my bank peaked at £770 and has since dropped in the last week down to £369. The main reason for this appears to just be the luck of the draw in that it's scores like 1-0 2-0 1-1 that are the winning one's and thus lower odds, rather than a nice 3-2 at 34 or something.

My strike rate up to the 8th Dec was 97 out of 177 54.80%. Since then it's been 31 out of 67 46.27% So, not a big drop in the strike rate really, especially when there's only been a third of the number of bets in the second sample as the first.

In that first batch there were two excellent wins of £201 and £209, but of course any game you miss the score means a £20 or more loss so it's not as if those two wins were making the system appear to work because the strike rate was good over the whole sample. I was particularly pleased with one which was Blackpool winning 3-0 away at 120/1 which my stats suggested should have been 28.31/1.

So I guess my question is am I looking at the monetary value and getting concerned the system may not work and wanted a fresh pair of eyes to look at the stats and get another opinion.

If you need to know anything else on the stats just say. I haven't recorded each games outlay stake unfortunately and I don't really want to go back over my profit/loss for 250 games to get it. Calculating all the losing bets the average stake was £19.24 so I would suggest we could estimate the average stake to be £20 which would give an ROI of (369.57 / (20*245) ) * 100 = 7.54% by my reckoning, so that is very encouraging.

I think perhaps I need to just keep going and it's just the initial profit was above where the trendline would be and it's 'correcting' itself now. After all value or not even that 0-3 if the true odds were as my stats suggested would only come in once every 28 games, so this is still a relatively small sample of games.

[mathare]

Anyway over all in this time there have been 245 bets. The stakes per game range from around £16 (8 value bets) to a high of £26 (13) but on average it tends to be about 10 value bets for an outlay of £20 per game. To date my profit stands at £369.57 which I'm really pleased with, however since starting on the 23rd November my bank peaked at £770 and has since dropped in the last week down to £369. The main reason for this appears to just be the luck of the draw in that it's scores like 1-0 2-0 1-1 that are the winning one's and thus lower odds, rather than a nice 3-2 at 34 or something. 245 matches (I assume you mean that rather than 245 individual bets based on the rest of your post) is a decent sample size so no worries on that front. How does this break down by league/country? Is any one league significantly more profitable than the others? Be careful with this though as splitting up the 245 matches will give smaller samples that will contain more outliers (or rather the outliers like the Blackpool 0-3 will have more significance on a smaller sample).


My strike rate up to the 8th Dec was 97 out of 177 54.80%. Since then it's been 31 out of 67 46.27% So, not a big drop in the strike rate really, especially when there's only been a third of the number of bets in the second sample as the first. I'm assuming you don't have enough data to be able to back test this as you need live odds so there is no way to tell what the long term SR is likely to settle down to. Hmmm.

For each match you have a number of bets on, what is the average odds of these for each match, and then the average of those averages? No, forget that, that's not going to give what I want. I don't think. I'm trying to work out what SR you need based on the odds. Your SR is around 50% so far, a bit over that in fact, so getting evens on each match (taking into account all the bets) would see you in profit. That means making around £40 on average per game - is that plausible given the odds? This is a tricky one as you're going to hit winners at 100+ in amongst some in single figures aren't you? Let me ponder a little longer...


So I guess my question is am I looking at the monetary value and getting concerned the system may not work and wanted a fresh pair of eyes to look at the stats and get another opinion. Without knowing what the SR and ROI could settle down to this is a tricky one to evaluate. I don't have much of a gut feel for it either I'm afraid. I do have several ideas though that may help at least one of us get a bit further with it.

You're backing to level stakes whenever you spot value - have you tried varying the staking based on the amount of value? Small bets on those prices that are just a bit of value and bigger stakes on the best value bets. And the other way round. I sometimes find with football that bets that seem to offer oodles of value include some information I haven't factored in - there is a reason for the odds to be that big in the first place. It's hard to say why that should apply here but worth considering all the same.

What sort of scores are you backing in each match? Is it generally the same set with a few added/dropped depending on the match in question? Are you always backing 1-0 (or whatever) for example? How profitable are individual scorelines? Again by breaking down the bets by scoreline you'll reduce the sample size but you may find that it seems that your model used to produce the value odds is underestimating the chances of a certain scoreline so you could consider dropping it from the bets. Your sample is too small to be significant for this sort of analysis at this stage but it is something to bear in mind as you accumulate more data.


I think perhaps I need to just keep going and it's just the initial profit was above where the trendline would be and it's 'correcting' itself now. After all value or not even that 0-3 if the true odds were as my stats suggested would only come in once every 28 games, so this is still a relatively small sample of games.I think you're right here, a stellar start is now seeing a correction back to something more akin to the norm but without more data it is really hard to tell.

If only there were some way of back-testing this, at least on one or two divisions. Talk me through how you run this system Matt. You have a model that outputs odds for each scoreline in matches and then what? Where do you get your odds and when? Do you just record correct score odds or match result odds too?

The bookies (certainly Will Hill and Ladbrokes) used to use a table of correct score odds based on the match odds. You'd read down the list of match odds and across to the correct score you wanted and the corresponding cell would contain the odds for that scoreline. It was printed on the back of the coupons but I don't know if they still have that as I haven't used a printed coupon for years now. If that is still in existence you could look up the match odds for the European games (using football-data.co.uk) and cross-reference those odds with the bookies' correct score odds tables and thus back test for those divisions. If you do this for the games you have Betfair odds for too then you can estimate Betfair odds for each of the scorelines too and calculate profits to both bookies' odds and Betfair prices.

[MattR]

Thanks for the detailed thoughts Mat, a couple of things you touched on crossed my mind earlier too.

Firstly the process I use is the same initial data that I use for my ht bets. It's the goals for and against for each team over the last 6 games broken down by each half for just home and away. IE The home team I only use their home stats and the away stats for the away team. Then I get an average goals per half for each team by using the average the home team scores + the away team concedes and vice versa for the away team.

This unfortunately doesn't lend itself to backtesting without an awful lot of work! These average goals per half are used for a set of correct scores for each half. Then for say a 2-0 final score I use the ht/ft combinations to get the odds, so 0-0/2-0 2-0/0-0 1-0/1-0 for a 2-0 FT score.

I haven't listed the individual scores bet on in each game but could back check on my pr/loss statement. Generally if the stats are suggesting a low scoring game then the lower scores are prevalent although if it's one sided it may just be 1-0 or 0-1 and not both and 2-0 but not 0-2 etc. Often similarly to how you found with your team ratings the away team odds appear to be generous more often. Like you, I've yet to decide whether this is a psychological bias on the part of bettors, or a genuine home team advantage that can't be factored into a set of past stats.

Oh and yes it is 245 matches not bets, so yes a fair sized sample.

If it's any help here are the frequency of winning scores to date. Oh and one thing I have noticed is 1-1 is rarely value. It's very often about 0.4-0.6 over the betfair price. There have been value bets on it but this tends to be when the home team is a bigger favourite. That is definitely the least bet score

Winning Scores
0-0 15 times
1-1 4
2-2 9
3-3 1

1-0 13
0-1 14

2-0 13
0-2 11

3-0 5
0-3 2

2-1 12
1-2 8

3-1 4
1-3 6

3-2 3
2-3 2

Any Other 9


It might be worth me getting the scores of those games that lost and see how that looks too. Would be interesting for a start to see how many 1-1's cropped up.


I had that idea about varying the stake according to the value as well, even to the extent of thinking what if each score had a £5 bet and then if it is perceived as non value then deduct a portion from the stake and add a portion onto the stake for those that are value depending on what percentage under or over. I'd have been fairly happy with Blackpool's 0-3 win using that system :yikes:

Obviously the biggest downfall with level stakes is that with an average stake of £20, that's £18 off any win right off the bat.

This is interesting
The average win is : £18.57 (132 wins)
The average loss is: £18.42 (113 losses) -missed out two days for some reason on my previous calculation of 19.24

So that suggests keeping the strike rate above 50% is the number one priority and that the average odds win has to be £36 less the other £18 of losing stake. So ave win odds of 18/1 are required. Hmm, that's not unreasonable I would think looking at the scores that have won. 3-2 and 2-3 are normally anywhere from 30 to 60, 3-3's 60 to anything, an underdog 0-2 can be 50+ and 0-1 around 20 too. So I would think the odd big odds win is required. Would you agree that would lend itself to the varied stake based on the value? That would make sense if 0-3 is for example coming up as 450/1 and it's 600/1 to bet, that would make it a smaller bet than if 0-1 is 20/1 predicted and 50/1 to bet, which would make a lot of sense to me.

I did also toy with the idea of just dutching all the value scores but I like the idea of picking up that big one every so often so am reluctant to do so.
Record by country

[mathare]

Firstly the process I use is the same initial data that I use for my ht bets. It's the goals for and against for each team over the last 6 games broken down by each half for just home and away.

This unfortunately doesn't lend itself to backtesting without an awful lot of work!Rubbish! :laugh A bit of VBA and Bob's your father's brother :wink


These average goals per half are used for a set of correct scores for each half. Then for say a 2-0 final score I use the ht/ft combinations to get the odds, so 0-0/2-0 2-0/0-0 1-0/1-0 for a 2-0 FT score.My god that sounds complicated. I understand what you're doing but not why necessarily. Why use HT goal averages and not FT averages instead? It doesn't matter to these bets when the goals are scored as long as they are scored before the ref blows the final whistle. I would question the validity (or more the point) of using separate first and second half data and combining the scores as you said. I'd calculate FT averages and Poisson those.


I haven't listed the individual scores bet on in each game but could back check on my pr/loss statement. Generally if the stats are suggesting a low scoring game then the lower scores are prevalent although if it's one sided it may just be 1-0 or 0-1 and not both and 2-0 but not 0-2 etc.No worries, I get what you're saying.


Often similarly to how you found with your team ratings the away team odds appear to be generous more often. Like you, I've yet to decide whether this is a psychological bias on the part of bettors, or a genuine home team advantage that can't be factored into a set of past stats. I am still pondering why the away odds are so great for some of my bets. It's something I hope to be able to analyse properly in the new year.


If it's any help here are the frequency of winning scores to date. So you're backing 0-0, 1-0, 0-1, 2-0, 0-2 and 2-1 most frequently. Small samples admittedly but that makes sense to me I think. 0-0 is generally underestimated by most punters and I'd wager that the low scoring bets are for games with a pretty strong favourite. It's interesting that on Betfair the prices are set by the general betting public rather than the bookies (OK, bookies lines move in response to the weight of money too but you know what I mean) so we could expect the exchanges to reflect public opinion largely - 0-0s don't happen that often (but more often than people reckon) and big favs will hammer their opponents (or will they?). Yes, interesting.


I had that idea about varying the stake according to the value as well, even to the extent of thinking what if each score had a £5 bet and then if it is perceived as non value then deduct a portion from the stake and add a portion onto the stake for those that are value depending on what percentage under or over. I'd have been fairly happy with Blackpool's 0-3 win using that system :yikes:I like that idea, that's good. Worth a look certainly. It's easy enough to run all sorts of staking ideas over the data once you have it thanks to Excel.


This is interesting
The average win is : £18.57 (132 wins)
The average loss is: £18.42 (113 losses) -missed out two days for some reason on my previous calculation of 19.24Now that is interesting, I agree. As you say it looks like maintaining that SR is important. The occasional 3-2 etc will throw up a big winner for you but you need the bread and butter of the lower odds winners too


Would you agree that would lend itself to the varied stake based on the value? That would make sense if 0-3 is for example coming up as 450/1 and it's 600/1 to bet, that would make it a smaller bet than if 0-1 is 20/1 predicted and 50/1 to bet, which would make a lot of sense to me. It's worth a try. As I said above with the data in Excel it should be a quick and easy job to test various staking plans. One that could be worth considering is the fixed profit (variable stakes) plan - stakes are varied with odds looking to win a set sum with each bet.

I'm truly up to my nuts in things I should be doing that aren't this but I have the feeling that a little Excel help from me could set you well on your way to having a real good stab at this and exploring it in a lot more detail. I might be able to help over the next couple of days if you'd like - I've probably written almost all the code you'd need at least once somewhere in my archive of workbooks.

[MattR]

Rubbish! :laugh A bit of VBA and Bob's your father's brother :wink .

:laugh




My god that sounds complicated. I understand what you're doing but not why necessarily. Why use HT goal averages and not FT averages instead? It doesn't matter to these bets when the goals are scored as long as they are scored before the ref blows the final whistle. I would question the validity (or more the point) of using separate first and second half data and combining the scores as you said. I'd calculate FT averages and Poisson those..

Well the thing is Mat, I'm using the ht scores for my ht bets so it was just as easy to split it into two. Actually it's been quite fascinating seeing how football really is a game of two halves! Some of the info is really useful to know and useful for betting decisions. For instance from tonights games Chelsea's first half stats are 7 for 2 against, for the second half 11-1. Montpellier with an excellent home record are 2-0 and then 8-2 for the second half.

I may be wrong here but my thinking is that if the distribution of the scores is so markedly different in a lot of teams cases then splitting it into two halves will throw out a different odds figure? For example above Montpellier to win 3-0 would seem most likely that they'd have to get at least 2 of their goals in the second half but what if their opponents haven't conceded a goal during the second half in the last 6 games - surely that would lead to a different outcome than using just a 10 for 2 against total? Actually, no I think I am wrong there?

Anyway as I say it's mainly because I started doing the half time bets and wanted to use just teams first half goals.





It's worth a try. As I said above with the data in Excel it should be a quick and easy job to test various staking plans. One that could be worth considering is the fixed profit (variable stakes) plan - stakes are varied with odds looking to win a set sum with each bet..

I like that idea too Mat, certainly worth checking out.




I'm truly up to my nuts in things I should be doing that aren't this but I have the feeling that a little Excel help from me could set you well on your way to having a real good stab at this and exploring it in a lot more detail. I might be able to help over the next couple of days if you'd like - I've probably written almost all the code you'd need at least once somewhere in my archive of workbooks.

That's really good of you Mat and very much appreciated. No worries if you can't though, don't put aside stuff you need to do on my account! :thumbs


Oh and I've attached the country record as a pic

[mathare]

Well the thing is Mat, I'm using the ht scores for my ht bets so it was just as easy to split it into two. Actually it's been quite fascinating seeing how football really is a game of two halves! Some of the info is really useful to know and useful for betting decisions. For instance from tonights games Chelsea's first half stats are 7 for 2 against, for the second half 11-1. Montpellier with an excellent home record are 2-0 and then 8-2 for the second half.I didn't realise the differences between the two halves were so marked, well well.


I may be wrong here but my thinking is that if the distribution of the scores is so markedly different in a lot of teams cases then splitting it into two halves will throw out a different odds figure? For example above Montpellier to win 3-0 would seem most likely that they'd have to get at least 2 of their goals in the second half but what if their opponents haven't conceded a goal during the second half in the last 6 games - surely that would lead to a different outcome than using just a 10 for 2 against total? Actually, no I think I am wrong there?Ooh, thinking head on now - with added furrowed brow as I am having to really puzzle this one out. Part of me is saying that Montpellier have scored 10 goals and conceded 2 so when they are scored is irrelevant. The point at which half-time is declared becomes irrelevant (I think). Why not look at goals for/against at 30 minutes and 60 minutes rather than 45 minutes? It's just that the game is played in two halves so data is recorded at HT and FT. Correct score betting is based on FT scores though so a large part of me is now thinking that HT data is irrelevant, as it data from any other time in the match. Maybe their opponents don't concede many in the second half but does that matter?

Yes, for Montpellier to win 3-0 you're looking at a higher likelihood of second half goals than first half goals so working out odds based on that and the knowledge of when their opponents concede will probably give a different set of odds from those calculated from using only FT goal information. But what I am trying to say is that HT is an artificial concept in many ways, an arbitrary division of the match into two approximately equal parts that has no effect on the settlement of the bet. Overall, more goals are scored in the second half of an average match for various reasons (including the fact it's usually slightly longer and there is more need for teams to chase a result than at HT) but why use data broken down into halves - why not into 10 minute intervals, 15 minute intervals etc? You're using data that happens to be available to create a set of odds when that data may not actually be relevant. That's not to say you shouldn't do it, of course. Your method may be as valid, if not more so, than my suggestion of just using FT for/against figures. It's not rocket science to run the two methods side by side in Excel either.


Anyway as I say it's mainly because I started doing the half time bets and wanted to use just teams first half goals.That makes sense, but as I said above for this idea I'd use FT data - it's certainly simpler.


That's really good of you Mat and very much appreciated. No worries if you can't though, don't put aside stuff you need to do on my account! :thumbsHave you got my email address? If so, send me a copy of whatever data you have in whatever format it comes in and I will see what I can do. I am out most of tomorrow but can free up time around that. I'd like to help as I am genuinely interested in this and it's not completely at odds with some other work I had planned actually. If you don't have my address I can let you have it

[mathare]

Oh and I've attached the country record as a picThe two countries that stand out for me here are Spain and Germany. The Spanish SR is under 50% but in decent profit - that suggests you've hit a big winner on that one. Keep your eye on this one to see how the bread and butter score bets go. For Germany the SR is almost the same as for Spain but you're showing a reasonable loss. Looks like this one needs one of the bigger bets to come in, else the low odds bets are actually too low to be able to show a profit. It's hard to say with any conviction what's really going on as the samples are too small but when you have more data this could be an interesting exercise to repeat. You may find it just doesn't work for certain countries and would be better off dropping them.

[MattR]

Yes I have your email address Mat, I'll send over what I'm using. I've just realised actually that this sheet I have I based from an old probability sheet I picked up on the internet and it uses a power function to calculate the probability. I did have one with poisson but I'm not sure where that went to then. It was in the same format which is why I thought it was a poisson formula in the box. I am pretty sure the two gave the same result anyway from when I checked them a couple of years ago. Anyway you'll know far better than me if things need to be changed there.

This is the formula in the cell

=(((POWER(H$83,X87))*POWER(2.718,-H$83))/FACT(X87))


where h83 is the home goals average. If i remember correctly the 2.718 was the average number of goals per game in the Premier league at the time - interesting. Going to alter that a bit and see what effect it has. X87 is the score for one team. I wonder if this could be deemed more accurate if you were to use the league's average goals that you were checking or whether just poisson is the way to go.

I'll send the sheet over so you can see for yourself.


Now I want to know where my poisson sheet got to :laugh That'll teach me for having them layed out the same!

[MattR]

The two countries that stand out for me here are Spain and Germany. The Spanish SR is under 50% but in decent profit - that suggests you've hit a big winner on that one. Keep your eye on this one to see how the bread and butter score bets go. For Germany the SR is almost the same as for Spain but you're showing a reasonable loss. Looks like this one needs one of the bigger bets to come in, else the low odds bets are actually too low to be able to show a profit. .


Yes that's exactly it Mat. Spain had a £200 winner. The German one's have had small winners bar one big win of £89. A lot of the wins on that have actually been minus as they didn't cover the other £18 stake. The other standout one like that is Romania which has a great strike rate but all of very small and minus wins bar one big one.


I'm going to check the ave goals per league online now. In theory if this power equation has merit then the leagues that are doing better should be nearer to the 2.71 ave goals - right?

[mathare]

I'm going to check the ave goals per league online now. In theory if this power equation has merit then the leagues that are doing better should be nearer to the 2.71 ave goals - right?That would make sense, yep :thumbs