Friday, 11 April 2014

Does this help to explain the issues this season?

I’m off work next week and ultimately, this is the week of the year when I start pulling together my plans and thoughts around what I need to do this Summer in preparation for next season. At the moment, I actually don’t have any concrete plans and it’s difficult to make plans when I’m feeling quite down about how the season has gone. I’ve already said the service needs a massive revamp, the ratings need a lot of work potentially, the systems need binned or at least, filtered in some way to reduce my workload and make it easier for followers of my ratings and systems. I can probably do with starting with a blank sheet of paper and mapping out how I think my ratings and systems should look but being honest, I don’t know if I can be bothered with it.

Anyone who reads my emails will know this season has been one massive pain the arse for me and when you have a month like Sep-13 was (massive losses), it suddenly seems like a much tougher season from there on and ultimately, just when I seemed to get back on an even keel, March then came along and was worse than September! At the moment, my mind is a little bit of a mush and it’s difficult to think clearly about where I go from here.  Tuesday night was like another massive blow (Sheff Wed and Woking losing late goals caused me to have a substantial swing on the night) and it’s hard to pick myself up at the moment.

Personally, coming into this season, I increased my betting bank by 150% from last season as most of you know and I had huge plans for how I would spend my profits this season! A new Garage for the house, paying off some more of my mortgage, reinvesting some more of the profits into my betting bank. That’s the 3 areas where my profits were going from this season. Ultimately, as things stand today, my betting bank is 20% lower than when I started the season.

I always knew my issue at this game would be when I had a loss. It’s easy to have the right mind-frame at gambling when you are winning but for someone who isn’t used to losing, how would I cope with a loss? Well, I already knew the answer to this and I’ve admitted it a few times but I’m a terrible loser. I hate losing (doesn’t everyone?) but I really struggle to accept losing. In the same way as in my personal life, I struggle to accept doing things wrong or struggle to accept criticism for the way I do things.

Now I’m facing up to a losing season, where do I go from here? Part of me says that I’ve had 3 great seasons, this season has brought me down to earth and now is the time to bail out and keep the money I’ve won over the last 4 seasons. I gave it a really good go, created something that was pretty good but at the end of the day, a season like this shows what can happen when things go wrong and ultimately, do I really want to risk losing money at betting?

At the moment, I’ll be honest and say this part of me is winning unfortunately. The other part of me are my usual characteristics and the fact I always believed I have something pretty special here and ultimately, I don’t want to stop after a losing season. I struggle to accept the fact I’ve made a loss this season (I shouldn’t lose at this game!) and therefore, I need to put it right and at the very least, get back the losses people have suffered this season along with me. I’m not a quitter!

I’m always honest on the blog and I know Steve (Daily 25) doesn’t really like his tipsters having emotions but thankfully, all my bets are automated and therefore, I can say what I want about the systems, my losses, how I feel etc. and it has no impact on the future performance of the systems.  Therefore, that’s where my head is at today. I need to quickly make some decisions as the greatest workload at this for me is during the Summer and if my heart (and head) are not in it, I can’t continue beyond this season. I’ll give it some thought next week and I’ll either start planning for next season and my Summer workload or I’ll start winding down.

One thing I have been giving a lot of thought to is the performance of the systems this season and the fact that the Draw Systems appear to be doing really well whereas the systems that are more established are struggling. Is this a new phenomenon? Well, no, because last year, the New Systems outperformed the Est Systems and therefore, is it the case there is a time decay issue with my ratings?

Before I try to answer this, I need to again state that I don’t refresh my rating algorithms every Summer with new data. If I want to do anything, I add new rating algorithms or add new systems but I don’t refresh the algorithms I already have. As I’ve admitted a lot over the last 2 seasons now, with hindsight, I’m not sure this is the correct way to approach this game but it’s the way I chose at the outset.

My concern with refreshing the algorithms was that it meant that every new season, I would effectively be following a rating algorithm that wasn’t the same algorithm as last year. The weightings will have all changed and therefore, I thought this brought additional risk. Obviously, it means that all my rating algorithms have a shelve life and I’ve made this clear since day one that the Est Systems would one day probably need to be replaced. Hence, I built New Systems a few years back. My intention was to continue building new algorithms each Summer but of course, I soon realised that there is a limit to what I could do and given the Est Systems continued to perform well, maybe I could squeeze a good number of years from each algorithm I currently had.

The other way to work with rating algorithms is to try to employ a method of continuous improvement with the ratings and every Summer, use the data from the season past, refresh the ratings based on the new data and then you are good to go for the new season. My issue with this and why I have steered away from this is that it seems so easy to fall into a process of backfitting data. Hence, I have an unbiased algorithm, I bring in new data, change the weightings with the various constraints on returns and bet numbers and then end up with a statistically biased algorithm.

After seeing the results this season, my thinking has gone 100% full circle and I think to stay ahead of the market, you need to be taking into account as much recent data as you have available. At the moment, my first two algorithms were developed using data from 2006 and 2009 and clearly, the power of these ratings have waned over the last couple of seasons. The second half of last season probably painted over the cracks with these two algorithms in particular as at Christmas time last season, I was all for refreshing these ratings in the Summer. The second half of the season was amazing though and I decided it wasn’t worth the risk.

What I have been looking at is the performance of my systems based on how long they have been live. At the moment, 40 of the 41 UK systems are in profit since they went live. The system that is not in profit is losing 0.2pts! As I’ve said before, I think my ability to build profitable football systems is probably unrivalled over the time I’ve been betting on football.

In total, the profits if following all 41 UK systems are:

A 6.5% return from nearly 27.5k system bets. Admittedly, it was a better at the start of the season but at the end of the day, when I started out at this game 4 seasons ago with my first system, if you had said I’d have these returns after nearly 4 seasons which included a terrible season in there, I’d have accepted this I’m sure.

However, the interesting point I want to make here is that if you break the returns down by how long the systems have been live, it looks very interesting. I currently have systems at the moment that are in their first season (Draws), second season (Some New Systems), Third Season (Some New Systems, Misc Systems, Under-Over Systems) and some systems in their 4th season (Est Systems).  Therefore, the Est Systems have been alive for 4 seasons but their results have got worse as the seasons have gone by. The same is true for the other systems in the main. Putting this altogether, you get the following table of results:

Now, I find this table absolutely fascinating! When systems are in their first season, as I already knew and as anyone following the systems will know, the returns have tended to be very good. Admittedly, this tends to be the time we go easy with the staking as the systems are unproven but as the results show, this is when the systems are at their most powerful. As time goes on, the systems and ratings appear to become less profitable. What causes this?

Well, I don’t think it’s too difficult to jump to the conclusion that due to the fact I don’t employ a continuous improvement process with my ratings and don’t use refreshed data, this potentially means my ratings become dated to an extent and therefore, the best time to catch my systems and ratings is in the first season. Of course, my advice before this season has always been to be wary of newer systems and ratings as it seemed risky to take the plunge with unproven systems when there were already proven systems available to follow!

I’m conscious of the fact that this could all be bullshi* and maybe I’m looking for a solution for a problem and it just so happens I appear to have found one but I actually think this makes sense. What it means though is that I have a big issue with my current ratings and systems as they are all out of date and need reworked!

This feels like this post has come full circle…..Do I have the appetite to start again and refresh the rating algorithms, build new systems and so on? I don’t know…………….


  1. Hi Graeme,

    I hope thing turn round for you. It is a difficult and I hope you feel more positive over the next few weeks.

    Best wishes,

  2. Cheers Jason.

    Been a very testing time this season but as always, it’s how you come out the other end that defines you as a person I think. It’s what I’m telling myself anyway!


  3. Dear god man. Please do get a grip.
    You've had a rough time for one season and you thinking of bailing out and letting the bookies off the hook ?
    Come on man at least give it another season to see what's what. If you don't in time you will regret not doing so methinks.
    Chin up soldier, shoulders back, chest out and go hunt some bookies profits me tiger !

  4. lol Shafto. My wife said a similar thing to me on Tuesday evening tbh after I said I had enough of betting on footie after Sheff Wed and Woking conceded equalisers! She actually said ‘grow up!’ lol

    Yes, part of me feels the exact same way mate and ultimately, I’m sure some people don’t like to read things like this on a betting blog – they’d prefer I bullshit about how this is a one-off and I’ll bounce back next season and make everyone millions!

    I am trying to rationalise the loss this season (tough season for the underlying results, ratings probably need refreshed, too many systems, not the best of luck with many bets) and part of me does think this season is maybe the coming together of lots of different things and ultimately, I’m paying for it all at once.

    There are other things going on in my life which are also not easy to think through and I do plan to lock myself into a room next week in my house for half a day to think this through and decide what the next steps are. I hope I can get my head around this and if I do, I have a helluva lot of work to do over the Summer and ultimately, I need to decide whether this is doable.

    Creating the best football ratings, systems and the best tipster service was very difficult and took an amazing amount of hard work and dedication. Trying to do it a second time would be even harder I suspect!

    Cheers for the comment although not sure if you are a subscriber or not Shafto…..


  5. Forcing yourself to not watch the game/result may help with the stress levels.

    Maybe you and Steve can setup a support group :)

  6. “ Therefore, the Est Systems have been alive for 4 seasons but their results have got worse as the seasons have gone by. The same is true for the other systems in the main. “

    This is not true.

    A quick check shows that the Est Systems has a significantly better second season compared to the first season.
    The same is true for the systems, that went live in 2011.

    What the table really shows is that the current season is losing.
    But did we not know this already?


  7. Anon,

    If I thought not watching the results would change the results, I can assure you I wouldn’t pay any attention to them! :)

    It’s a valid point but I actually enjoy the ups and downs although admittedly, there have been more downs than ups this season. I’m sure Steve would say the same. At the end of the day, this is a hobby. After spending the time I spend on this, watching the results is the fun bit.


  8. Hi Dmitri.

    I wouldn’t say the Est Systems had a significantly better 2nd season than 1st season. The 1st season was amazing, it was ruined by two crazy months with draws. The second season had 9 months profits in a row as you know. In terms of returns, the first season was 9.4% (even after two crazy months) and the second season was 11%. The 3rd season was 8.1%. I agree there isn’t much in it until the 4th season.

    The New Systems (41,42 and the 6 combined systems) are in their 2nd season , their first season was last season. Systems 31,32 and 33 were live the previous season though, so have an extra season.

    I think you are correct though in the sense if I remove this season’s results (including the draw systems), the results are:

    Overall 9.1%
    Season One 8.2%
    Season Two 10.9%
    Season Three 8.1%

    Not many conclusions you could have drawn coming into this season!

    I guess my own view and this comes from my own experience is that the Est Systems have been fading although last season was a strange one as they were on their knees at Christmas but rebounded very well and ended up with a 3rd great season! The New Systems were unreal last season but even when things have been going well this season, I’ve never quite experienced the same returns as last season.

    STOZ has dropped off a cliff this season, looked like an amazing system last season!

    As a matter of interest, do you think footie ratings should be refreshed every Summer or should they be left alone and then be retired after X years and replaced by completely new ratings each Summer?

    Is this season a blip in your opinion? Do the underlying results explain the poor performance this season or do you think something else is at play?

    Happy as always to hear opinions from yourself or anyone else reading. The more opinions, the better!


  9. You should have a chat with Jonathan Lowe from Steves favourite service Sportpunter. Jon has made ratings for AFL and tennis for a decade now so should know a thing or two about refreshing them or not.

  10. Every bettor following two different tipsters or systems has probably experienced the following:
    In a game between teamA and teamB one tipster selects teamA, the other tipster selects teamB.
    The bettor must decide how she will bet this game. But the information can help if used wisely.

    For TFA UK there are 4 independent ‘tipsters’ called Algorithm 1 to Algorithm 4.
    But the strange thing is, that subscribers are not fully informed of the selections.
    When two or more algorithms disagree, Graeme, not the subscriber, decides how to bet such a game. You will never see opposing teams on systems 6, 21, 31 and 41.
    Therefore these 4 systems are dependent.

    System 6 is different from a season to the next. In the first season there were only two algorithms, the next season three and the next season four. So to compare ‘system 6’ over four seasons is incorrect.

    It should be noted that for the Euro bets, Graeme selects both teams when the two algorithms disagree.


  11. Hi Anon.

    I have contacted Jono and have some things in the pipeline this Summer with others too. I’m always keen to learn at this game and therefore, I’m always happy to seek help where possible. More so now than ever after this season!



  12. Hi Dmitri.

    I agree to an extent with what you say but there are only a small handful of conflicting bets a season on the UK algorithms. Therefore, it is not material. I changed the way I deal with conflicting bets on the Euro bets this season and I don’t think it has made too much of a difference. There are such a small number of conflicting bets, it can’t make a difference.

    With hindsight, I should have done this with the UK bets to keep things cleaner but seeing conflicting bets for the same games would have confused some people! Now people understand my algorithms and systems better, seeing conflicting bets wouldn’t be such a big issue although to an outsider, a tipster tipping Home, Away and possibly Draw in the same game would look strange. :)

    Cheers for the comment.


  13. Hi Graeme

    Thanks for the answer. To be honest I ‘think’ you are playing the guessing game. Why not present the actual numbers? That would make it possible to give useful feedback.

    E1 and E6 have 18 conflicting bets so far this season.
    Not that many I agree, but hardly surprising when you look at the average odds for the two systems, 3.53 and 3.70. They play on the same field.

    This is similar to season 2010-11 when system 6 and system 21 had no competition from other algorithms. Average odds were 2.71 and 2.73. 65% and 67% of all bets were at odds < 3.00, so these two systems were obviously build to favour the lower odds.
    This would actually be very good indeed, because with systems 31 and systems 41 coming in the next two seasons and focusing on longshots, TFA had a perfect mix that allowed customers to diversify their portfolio.

    So in season 2011-12 system 31 were introduced with average odds of 4.02.
    Now one could think that two systems preferring low odds and a system preferring high odds would result in more than a small handful of conflicting bets.
    I can only guess of course as I don’t have access to the actual facts, but we can all see in the published tables, that average odds for system 6 went from 2.71 to 3.68 and average odds for system 21 went from 2.73 to 3.89. And now only 38% and 30% of all bets for systems 6 and 21 were at odds < 3.00 indicating that a fair number of bets at low odds are filtered away.

    I find it fascinating when system 21 in season 2010-11 had 326 out of 489 bets at odds < 3.00 the next season only had 169 out of 562 bets at odds < 3.00.
    I’m sure a statistician would be able to show that a drop from 67% to 30% could not be the result of chance alone. But the only change between the two first seasons is the introduction of system 31 and by that more conflicting bets.
    But you say that conflicting bets are not material. Do you have another explanation for the surprising numbers?

    Anyway, now we are left with 4 longshot systems which makes us vulnerable. We are not able to diversify using only TFA service. We have to use other services, just as you do yourself, to get the protection.


  14. Hi Dmitri.

    The decision to remove some of the lower priced bets from the first two algorithms was made after the first season. An additional filter was applied the season after too, thus removing all lower priced bets from these algorithms and bringing them more into line with the 3rd/4th algorithms.

    I think my thinking is getting closer to yours now and that by removing these shorter priced teams, I’ve increased the variance a great deal.

    These bets were loss making in season one , break-even in season two and loss making again last season. However, this season, these bets are 10% ROI and ultimately, in a season when short prices have done amazingly well, it would have recouped the losses from last season and much more.

    I think with hindsight, I should have settled for a lower ROI and keeping segments of bets that were unprofitable. I think the 3 seasons before this season were very favourable for Aways and therefore, I probably forgot about what happens if the balance swings towards Homes. Well, I have now seen what happens, my ratings and systems really struggle!

    What I need is a better balance I think.



  15. I wrote in an earlier post:
    “Every bettor following two different tipsters or systems has probably experienced the following:
    In a game between teamA and teamB one tipster selects teamA, the other tipster selects teamB.
    The bettor must decide how she will bet this game. But the information can help if used wisely.”

    The keyword here is ‘wisely’.
    Let’s take an example:
    One tipster selects TeamA as AH-0.5 at odds 1.90.
    The other tipster selects TeamB as AH+0.5 at odds 1.90.
    This is a case of conflicting bets. Betting both selections will obviously result in a certain loss of 0.10 units. This is not ‘wisely’

    Many bettors are of the opinion, that if following two systems, the optimal strategy is to simply place one unit on each selection, conflicting or not.
    Given my example I think that Graeme and Steve from Daily25 will realize, that such a strategy is not optimal.
    The conclusion in my example is so obvious because the bets are in conflict.
    More often the selections are in harmony, both tipsters select the same team.
    In this case it must be optimal to bet two units on the team. Right?
    The logical consequence would be, that if 100 tipsters select the same team, you should place 100 units on this team, hardly wisely!

    When I mentioned Graeme and Steve by name it’s because we all know them and they use this strategy.
    In my opinion, and I could be wrong of course, this is the main explanation for the poor results this season.
    Steve is writing an ebook about betting. It will be interesting to read, what he has to say about this important topic.

    Maybe Graeme and/or Steve would come in here and explain why I’m wrong.


  16. Hi Dmitri.

    After this season, I agree with your thinking 100%. Hard to disagree after seeing the mess Steve has made this season with his portfolio and the loss I’m currently sitting on. As always though, it’s easy with hindsight to look back at things and think there are better ways to do things. :)

    Everyone following correlated portfolios and lumping on certain teams were working on the assumption that the ratings would all be highly profitable again. Hence, filtering systems and bets when my ratings are doing well isn’t that difficult as we saw in the first 3 seasons and ultimately, it was difficult to lose!

    The issue that I have encountered and the majority of subscribers with the service this season is that I honestly didn’t think that my ratings could have a losing season. I would have bet big money systems 6,21, 31 and 41 would all achieve 3%+ ROI this season and ideally, 5%+. Hence, to see 3 of the 4 rating algorithms losing is basically the first issue. Once you start staking more and more on teams that appear on these systems, you are just compounding losses unfortunately and as Steve has seen, compounding losses can really hurt!

    The issue this season IMO is nothing to do with how people have played the bets. As I’ve said before, over the first 3 seasons, it was impossible to lose. People could have done anything and make money. People entered into this season believing that this season would follow the last 3 seasons. Hence, it would be impossible to lose. This season has shown that if things go against you (underlying results have been difficult and we have had more our share of ‘bad beats’ for want of a better phrase), it can make it very difficult to make a profit.

    What we all need to do is adjust our expectations based on what has happened this season. IMO, losing 2%-3% with the ratings isn’t a disaster season. What people have done with the ratings and systems and managed to lose 50%+ of betting banks is the disaster this season!

    I’ve said this to people before who have contacted me and shared advice Dmitri but I’m always open to people wanting to help others out. Hence, if you want to tell people which bets to follow to make a profit with my ratings, feel free to do so. :) What I find though is that people aren’t willing to put their necks on the line and therefore, it’s easy to pick holes in what others do rather than suggest what they should be doing! Why don’t you tell Steve what bets to follow next season?

    Same comment to anyone reading this. My blog is open to anyone who wants to pen an article or two on how we should use the ratings and systems. Just contact me.

    Cheers for the comment again. :)


  17. Hi Graeme

    My last post was a reaction on your email about the future of the service.
    I point to some errors made in the past, of course not to criticize, but to give constructive help so to avoid these mistakes in the future.

    Three things in the email stood out.
    There will be a significant change in the average odds, there will be conflicting bets a la Euro algorithm and there will be cross referring of algorithms.

    After lowering the average odds on f. ex. algorithm 1 it’s obvious conflicting bets will be an issue.
    Therefore I showed that it will not be the best solution to automatically advise conflicting bets.
    My second point was about how to bet, or rather not to bet, a team appearing on multiple systems in a portfolio.

    I follow Tage Poulsen on Twitter. He wrote a note early in the season to all your subscribers (incl. Steve!) where he warned about betting too much on multiple system picks. And he presented a complete system, called Max4, based on systems 6, 21, 31 and 41.
    He tweeted this morning and told Max4 is up 97 units (6.4% yield) this season.


  18. I saw the tweet from Tage but I could hardly retweet that! Given how much stick I’ve had this season from the fact people have had such various results, I’m not going to advertise a way of following which wasn’t available at the start of the season. Not only that, I know two people who started following soon after I sent the method from Tage and it then went on a terrible drawdown and they quit before it bounced back! I can never please everyone at this game. :(

    I guess what I’m asking for is your opinion of what I should do next Dmitri. It’s easy to say I shouldn’t now include shorter priced bets as we both know all that does is reduce the likely return in the long-run but it would potentially reduce the likelihood of another season like this one being so extreme.

    Which rating algorithms would you keep if I said I could only keep two? Would you filter the algorithm at all? What other systems would you like to see? What do you make of the Draw systems so far this season?

    At the moment, the service really is a blank sheet of paper and I’m happy to hear suggestions about how I move forward. Looking at the live results of the systems and algorithms, it’s not easy. I look at a system like 8-22 which is having a shocking season but has a 15.5% ROI since inception. Should I consign this to the scrapheap just because it doesn’t have enough bets? I know one person who follows this in isolation and has made amazing profits since inception and only wants 30-40 bets a season. Do I retire this system and lose his membership?

    Put yourself in my shoes. What is the way forward for the service in your eyes?


  19. My best advice to you would be to stop punting a year or three and concentrate on your daughter, your wife and your new job!

    “It’s easy to say I shouldn’t now include shorter priced bets as we both know all that does is reduce the likely return in the long-run”

    I never said that. On the contrary I’m a strong believer in short priced bets. What I said, is that the conflicting bets should not be handled the way you say in the email.
    It’s correct that betting on shorter prices will reduce the likely return in the long run. But only if you use level stake betting. That’s the reason why you bet two units on certain home games.


  20. Hi Graeme

    Apropos my ’best’ advice, I got this tweet:
    “Roger Federer may have to miss the French Open next month because he wants to be with his wife when she gives birth to their third child.”

    I have thought very much about your problems with the service going forward.
    Given that the service at the moment really is a blank sheet of paper, you might be interested in my suggestions.
    To explain myself I unfortunately have to be a little technical and the post will be longer than usual for me.

    The term INDEPENDENCE is used in probability theory and has a strict definition.
    It’s easy to get it wrong because if I say:
    “The value of this pick DEPENDS on the bookies odds” I use the word DEPEND in the way we usually do when we speak or write, not in a probabilistic sense.
    Betting is very much a matter of using probabilities to quantify the chances of random events, f. ex. a football game to be played tomorrow.
    Example A. Tottenham – Fulham in EPL and Osasuna – Valencia in La Liga.
    Both games are random events. Are the two games INDEPENDENT?
    Yes, because the probability for Valencia winning will not change even after the first match has finished hours before the second starts.
    Example B. The match Oxford – York today is a pick on System 31 and D1.
    Are the two picks INDEPENDENT?
    No, because if you know that D1 is winning you also know that the other pick is losing.
    Example C. You have two bets in the game Tottenham – Fulham. Over 2.5 goals and Tottenham winning 3-1.
    Are the two bets INDEPENDENT?
    No, when Tottenham is winning 2-1 the first bet is a winner, but you still don’t know about the other bet.
    Other examples:
    System 7 and System 8. DEPENDENT
    System 21 and System 32 DEPENDENT
    TFA and Football Investor DEPENDENT
    Football tipster and Tennis tipster INDEPENDENT.

    When two systems or tipsters are INDEPENDENT it’s easy. You can just place all picks in the normal way.
    In case of DEPENDENT systems or tipsters it’s much more complicated. And with a staking strategy of level staking, betting one unit on all picks, it’s simply impossible.

    A few days ago Steve illustrated the problems in his blog:
    “The Skeeve bet was opposed to a big TFA bet on Braintree and I had to make a hard decision on whether to place both bets, to this day I have never made the right decision, but decided to treat each model independently, and placed the bet, TFA won that bet and it was a crazy game. TFA made $3,100 thanks largely to that one winning bet.”

    Steve was indeed lucky in this case.
    He made two blatant errors and got away with them.
    First he treated Skeeve and TFA as INDEPENDENT, which they aren’t.
    Second the TFA bet was ‘big’ meaning he treated all the systems that picked Braintree as INDEPENDENT, which they aren’t.

    My suggestion.
    ALL systems shall be INDEPENDENT.
    In that case a subscriber can use multiple systems in the normal way without problems.

    That will put severe restrictions on you. You can only have one UK algorithm. It’s no good to have a HOME system by using a filter on algorithm 1 and an AWAY system by using a filter on algorithm 2.
    You may think that a HOME system and an AWAY system build this way are INDEPENDENT. They are not, because the bets come from two DEPENDENT algorithms.
    But you could have:
    System Low odds, system intermediate odds and system high odds.
    Or system EPL and system not EPL
    Or system Home and system Away (and system Draw if the algorithm has prices on all three 1X2 possibilities)
    And then of course some Euro systems which come from an algorithm that is INDEPENDENT of the UK algorithm.


  21. Hi Dmitri.

    Apologies for the delay in responding, was spending time with my wife this weekend getting baby stuff. If I honestly thought my punting or the service impacted my family life, I wouldn’t be playing this game. This is a hobby to me and will always be that. I don’t need to be doing this and I do it for fun even though at times, it irritates the hell out of me! That’s part of the fun though. :)

    I’m not sure I agree with the fact that I can only have systems which are independent of one another. What I do agree with though is that people need to appreciate the fact that systems aren’t independent and therefore, putting correlated systems in a portfolio isn’t the best way to use the systems as I’ve seen first hand this season.

    As you probably know, when I set out, the intention was never to have people following systems in a portfolio. My initial thinking was that people would choose a system or algorithm to follow and therefore, we all follow what we want. However, people then realised (as I did!) that by combining correlated systems, it could give you very high returns for very high risk and this is what attracted many people to TFA. Making 200%+ return on capital in 9 months if things went really well became possible although the unknown was always going to be what happened when things didn’t go well.

    I think this season has opened the eyes of everyone following the service about what happens if things go badly. That’s not a bad thing for the long-run.

    I guess your solution leads me down the avenue of one algorithm and possibly systems built from this algorithm. As soon as I start having different algorithms, it becomes more difficult. This is one solution but as I’m sure you are aware, there are many other solutions to this problem I suspect. I just need to try to find the right one for as many people as possible, including myself.

    As always, thanks for the comments.

    I’m probably going to hold off until season end before mapping out the future of the service as quite simply, I don’t see an obvious solution. Having one algorithm appeals to me but what it then means is that all my eggs are in one basket. As I’ve said before, I have no intention of following other tipsters beyond what I currently follow and therefore, relying on one algorithm is risky for me. If I build that algorithm wrongly, I’m screwed along with everyone in the service!


  22. This post is written exclusively for Steve, so if your name is not Steve and you still go on reading, then I take no responsibility!

    For a start Steve, every bet you place at Sportpunter has the stake calculated by the Kelly Criterion.

    As you know the formula is: Kelly Stake = Edge/(Decimal odds -1)

    Let’s assume your bank is $10.000

    Example A. Edge = 5%, Odds = 2.00. Then Stake = 0.05/(2-1) = 0.05 = 5% = $500

    Example B. Edge = 5%, Odds = 5.00. Then Stake = 0.05/(5-1) = 0.0125 = 1.25% = $125

    So at all your Sportpunter models you are using variable staking.

    At TFA and the other football models you are using level staking. Let’s assume one unit = $200.

    Therefore if the situation in example A occurs in your TFA model as a single bet, you will stake $200.

    The Kelly Stake of $500 is the mathematically optimal bet size. So in this case you are underbetting and that will result in a lower profit in the long run, but still a profit if your edge is 5%.

    In example B you are overbetting. This will also result in a lower profit in the long run.


    It’s also a mathematical fact, that a stake greater than 2 * Kelly Stake will result in a LOSS in the long run.

    You should be scared now, because when you have one of those ‘big’ multibets of say 8 units, you are risking $1600 in example A, much more than the limit of $1000. The result will sooner rather than later be an even greater number after the minus sign at TFA model.

    And think about it. You have many different football models, so you have many situations of LOSING multibets, even though you have a positive edge.

    At Sportpunter you have generally only one model per sport, so you avoid multibets.


  23. To Steve, from theory to practise.

    As we saw, mathematical theorems tells us, that placing bets greater than 2 * Kelly Stake will lose you money in the long run
    Of course you can be lucky and actually win money in a shorter period, but that luck will not last.

    With your TFA models you don’t know what the Kelly Stake is for an individual bet. That would require Graeme to inform us of the Edge.

    But from the two examples I showed you, I would say that placing up to three units would be relatively safe.

    I have made some calculations on your TFA models performance in this season.
    You have 13 models of which two are draw models. So I will use the 11 non-draw models.

    Until now (April 21 noon) there have been 495 unique selections.
    303 of these were ‘small’ , ‘safe’ bets with 1-3 units.
    192 were ‘big’, bets with 4-11 units.

    The theory says that the ‘small’ bets should do better than the ‘big’ bets.
    And in this case theory is confirmed.

    The small bets has made a profit of 46 units.
    The big bets has made a loss of 99 units.

    This is the results of the past. What the future will bring you, nobody can tell. But you need luck to profit from your method of playing multibets.

    I assume you would never bet 2 * Kelly Stake on a Sportpunter model.
    Because you trust the Kelly Stake to be optimal.
    And it is. That was shown by Kelly in 1956 in a scientific journal.
    He also showed that it would be loss making to bet more than 2 * Kelly Stake.
    I think you should trust this theorem also.


  24. A useful benchmark for the LUCK factor in betting.

    A bet placed at fair odds 4.00 has 25% of winning. Just calculate ¼ = 0.25 =25%.
    Therefore after making 40 such bets your average number of winners is 40 * 0.25 =10.

    This number, 10, is the benchmark for this betting record of 40 identical bets.
    If you actually had 12 winners, you have beaten the benchmark, and the interesting question would then be, if you were lucky or clever.
    To answer this question other methods has to be used. But knowing the benchmark number is useful.

    If you have a betting record with different odds, as we usually have, just calculate 1/odds for each bet and then sum over all bets in the record. It’s easy to do using a spreadsheet.

    To illustrate this I take a betting record of all TFA multibets in algorithm 1-4, where the difference between winning and losing the bet is more than 50 units.
    The current season has 36 bets in this category of which 8 are winners.
    The benchmark is 8.3.
    In this sense the season has hardly been a ‘disaster’ (Graemes word) for the big bets.
    The 36 bets actually made a profit of 13.6 units.
    So a season close to expectation, maybe a little better when we consider that multibets are losing bets in the long run.

    Let’s compare to last season, where TFA had the exact same systems.
    There were 46 bets in all of which 16 won. The benchmark is 11.2. So nearly 5 more winners than the benchmark.
    What about the PL.
    Not bad, not at all. Profit = 401 units.
    My conclusion for last season is, that we were very lucky. 43% more winners than the benchmark!



    I forgot that my spreadsheet use fractional UK odds format instead of decimal odds.
    The two illustrations for multibets are therefore not correct.

    I will not show my numbers now, I’m in shock!!
    The new numbers are so unbelievable, that I expect to find another error in my computations.


  26. Well, a few drinks have definitely helped me. So I’m ready to make a new attempt.
    If you don’t like numbers (I do!!) please stop reading now. NOW I said.

    In this post I will show how algorithm 1-4 have done in general.

    We have data for 4 seasons (Vivaldi)
    First season 2010-11 had only two algorithms.
    In second season algorithm 3 was added.
    And in the last two seasons there has been four algorithms.

    This is reflected in the number of unique bets and how big the multiple bets can be.

    I use the following codes:
    N = number of unique bets
    AW = number of Actual winners
    EW = number of Expected winners
    PL = total profit for the betting record

    2010-11 N = 763 AW = 313 EW = 308 PL = 243
    2011-12 N = 946 AW = 305 EW = 289 PL = 406
    2012-13 N = 1134 AW = 394 EW = 377 PL = 750
    2013-14 N = 1026 AW = 341 EW = 327 PL = 50

    You will see that there were relatively many winners in 2010-11. But that is because Graeme had many low prices this season. He changed the first two algorithms next season completely by removing all these selections.

    Apart from that the last three seasons are characterized by having approximately 5% more winners than expected. This would indicate that the algorithms have an edge of about 5%.
    BUT that is only true if all bets are played for one unit each.

    By using multibet staking it’s important WHICH unique bets are winners.
    It makes a difference in PL to win an odds 8.00 bet for one unit compared to the same selection bet for 25 units. As the more math minded persons will realize, the difference in this case is 168 units!

    Therefore we must expect to observe great variation in the PL measured over the seasons. And that is indeed what we see, with this season being the worst.

    In the next post (tomorrow) I will examine the connection between PL and multibets,


  27. I have decided to make some reflections on TFA model before I present more detailed analysis.
    As usual I try to be objective. So I’m not negative when I disagree with Graeme.

    I will mostly concentrate on the last two seasons because they have the exact same structure, so it makes more sense to compare them.

    We are studying a portfolio of 25 different systems taken from 4 different algorithms.
    All 25 systems depend on each other, so it’s possible to see the same selection on all 25 systems, making this a multibet with a stake of 25 units.
    If the 25 systems were totally independent ALL bets would be single bets. You would NEVER see the same selection on two different systems.
    This is really two radically different ways to build a model.
    Graeme has chosen the first design intentionally. As he writes on the website:

    “The Football Analyst is the ONLY service in the market using multiple ratings to create combined systems and is why The Football Analyst systems are the best in the market by a long way.”

    This philosophy is based on a simple, intuitive idea, that if 8 systems agree, then it must be correct to stake 8 units on the selection. Graeme usually calls the greatest multibets for the ‘better’ selections.
    The philosophy of the independent model is based on the belief, that a selection has a fixed, unknown probability of winning. What tipster1 or tipster2 believe is irrelevant, so this model will use the same stake always if using level stakes as TFA does.

    To help us evaluate the two models we have two important results by Kelly.
    The first result tells us, that if we know the odds (as we do) and the probability of selection winning (this is unknown so we use an estimate) it’s possible via a simple formula to get the optimal bet size.
    The second result tells us, that if we stake more than 2 times the optimal bet size we will lose money in the long run.

    These results can’t be disputed, they are mathematically facts.
    But the Kelly results can help us.
    Say we use £10 as our unit stake and a selection has £25 as the optimal bet size.
    Then a single bet of £10 is underbetting and you will not get the optimal payout.
    If exactly two systems has the selection, then model1 requires a stake of £20. This is closer to the optimal bet size, so in this case model1 outperforms model2. £30 is actually as good as £20, they are both £5 from the optimum.
    Staking £80 will be loss making, it’s more than £50, double of optimum, so in this case model2 is better than model1.

    We can now make two conclusions.
    If we can estimate the win probability fairly accurately, the best model is to use variable staking (like Sportpunter) using a Kelly model.
    If we prefer to use level staking (like TFA) it’s impossible to stake the optimal bet size.
    In some cases model1 is best and in other cases model2 is best, so we see, that it’s not a choice between the two models. A hybrid model is better.

    In my next post I will use the data from the last two seasons to try and get closer to this hybrid model.


  28. We have seen, that our portfolio last year had a benchmark EW of 377 and actually beat this number by 17.
    This season the benchmark EW is 327 so far (the season is still running) and we have beaten it by 14.

    This model1 has been identical for two seasons and has been able to find winners at the same rate both seasons. This more than indicates, that nothing has changed regarding the basic algorithms.

    And yet we can state, that profit last season was a massive 750 units and this season a mediocre 50 units.
    Only one explanation is possible. Multibets.

    So let’s take a look at the biggest multibets.
    A 20 unit bet at odds 4.50 will return 70 units if winning and -20 units if losing. So there will be a difference of 90 units in the account depending on the outcome of this bet.

    Taking all bets where the difference is more than 80 units I find 20 bets last season with 6 winners and 15 bets this season with 3 winners. The average difference is 90 units.

    We have 2160 unique bets total and are looking at the 35 biggest bets. That is we look at only 1.6% of all betting opportunities for two years.
    Every time we change a winner to a loser or a loser to a winner in this small sample the account will change with 90 units.
    Imagine that we reversed the numbers so there were 3 winners last season and 6 winners this season. Then the profit last season would be 480 units and this season 320 units.

    It’s obvious that the fate of model1 is closely linked to the biggest multibets. And you can’t draw any meaningful conclusions based on 35 bets out of 2160. Here we are really talking luck.

    So when Graeme time and time again reminds us of the great historical results for model1, it’s based on hope rather than facts.

    A model should not be evaluated by ££.
    If so, any lotto winner is using a great model.

    To this comes an often forgotten ‘tipster bias’.
    You will practically never see a tipster with a negative first season.
    He would have to close shop because his customers are leaving searching for the next lucky tipster.
    Therefore a ‘live’ tipster will always have a great historical record.

    The trick is still to decide if this is due to luck or skill.

    Kelly rule two told us that these big multibets are losing.
    So whatever Graeme decides to do for the coming season, I think it would be ‘unwise’ to allow these bets in the model.

    Now that these bets have been removed we are getting closer to the hybrid model.

    To be continued.


  29. Hey Dmitri,
    Steve here from Daily25, I was directed here by Graeme, I usually never read comments so would have missed these normally.
    Firstly my maths skills are nowhere near yours or Grames. While I can understand what your saying alot still goes over my head. That's why I pay others to come up with the bets.
    You make a good point in regards to Kelly betting and it does seem to hold true that betting the 2x is what has cost me this season. I would never dream of betting full kelly and I was doing worse then that with these bets.
    I don't know if there is a correct answer on what to do when two models have competing bets, I'm sure there might be a theoretical maths answer, but I would still not know what to do. If I treat each model as independent then I will lose whatever the bookies edge is by betting both sides, I don't know which bet is the one with the edge and should just not bet any of them. I'll look into this much further when writing the book.

    I don't think the book will hold any appeal to you as it will not come at the problem from a maths based view, as mentioned my maths is at best adequate and the book is going to be more for the novice who is looking to make a second income. They wouldn't understand and I wouldn't want to confuse them with too much maths. I will off course make sure they know the basics, but anything above that is for the likes of you and the tipsters to discuss.

    Happy to talk to you further over email at admin @ mysite .com


  30. Hi all,

    I agree with Dmitri in a few things, and disagree in others.

    "My best advice to you would be to stop punting a year or three and concentrate on your daughter, your wife and your new job!"

    Completely disagree! Please, Grame, do not keep in mind that advise!

    "More often the selections are in harmony, both tipsters select the same team.
    In this case it must be optimal to bet two units on the team. Right?

    I agree.

    "My suggestion.
    ALL systems shall be INDEPENDENT.
    In that case a subscriber can use multiple systems in the normal way without problems.
    That will put severe restrictions on you. You can only have one UK algorithm"

    I disagree. There is no problem in systems being dependent, on the contrary, having several algorithms gives robustness and is positive. You just have to take it into account and to stake cleverly when the same bet appears in more than one system.

    "It’s also a mathematical fact, that a stake greater than 2 * Kelly Stake will result in a LOSS in the long run.
    You should be scared now, because when you have one of those ‘big’ multibets of say 8 units, you are risking $1600 in example A, much more than the limit of $1000. The result will sooner rather than later be an even greater number after the minus sign at TFA model."

    I agree, and I have said it in another post. If you stake one point for each time a team appears in a system then you are overstaking and you are going into bankruptcy in the long term, no matter the edge the system has.

    Graeme, you've done the hardest part, developing a set of systems with an edge, but you are failing with your staking plan, the easiest part.

    Regarding the service, I like as it is. Having so many systems can be a little confusing for some followers, but for me it's the beauty of this service, it makes me feel that I have some participation in my results, and I am not merely backing a list of teams at given stakes.

    To improve the results I definitely think you should refresh the ratings each summer , and it could be enough. The service is still great, this is my first season and I have an ROI of 0.6%, but it could be around 3% if I did stake the draw systems as much as I staked the proven systems. It's not as spectacular as the past seasons, but this season was rough in general, and a small profit is still a profit. Considering the high odds, a season like this can perfectly be due to variance; you have to see the whole picture, the four seasons.


  31. Hi Steve,

    I’m sure many bettors like you pay a number of football tipsters to come up with the bets.
    And when your ebook is finished this number will only get bigger.

    This will put you in a dilemma, that simply can’t be resolved.
    Take any of your football tipsters.
    Does any of these people you pay money to inform you of the edge on their selections?
    I know personally only one football tipster who supply this vital information.
    (I don’t understand how a tipster can tell me that teamA has value, that is has positive edge, but he is unable to quantify this edge. It smells like he is only guessing that teamA has positive edge)

    Without knowing the edge you can’t calculate the optimal Kelly stake.
    So all these tipsters use flat betting, they are practically forced to it.
    And by doing this their subscribers will get lower profit unfortunately.

    Take one of your most successful Sportpunter models and calculate the yield using level stake betting. Compare with the actual yield from using Kelly staking. I think you will be surprised!
    A main reason that you have most profit at Sportpunter is NOT that they come up with better bets than f.ex. TFA, but that you can stake optimally. It’s really a staking problem, not a selection problem.

    Now if five of your football tipsters agree on a selection, you will happily place one unit at all five tipsters. So you are in reality making a multibet. And Kelly rule 2 will probably be violated. You have just made a loss making bet!
    And if your tipsters have opposing bets you don’t know how to tackle this unpleasant situation.

    So when you sell the ebook to Joe the Punter he will subscribe to 8 different football models and he will not understand why he is losing after a year. But he will every weekend be guessing what to do with conflicting bets. And you can’t help him.

    In an ideal world every tipster supply you with the edge.
    Then every betting situation can be calculated and the optimal stakes determined.
    But you could not write about the solution in the ebook. It requires math at a rather high level to understand.
    But once you understand it, you are able to bet optimally in a golf tournament with 150 players and you have found value in 5 different players to win the tournament.


  32. Last post on analysis of TFA model.

    A The actual TFA model uses level staking and has many multibets.
    B The independent model uses level staking and has only single bets
    C If we knew the edge a Kelly model could be used.

    Model A is dangerous. A few bets and luck will be responsible for the result as we have seen in the last two seasons. Many bets will be unprofitable because they violate Kelly rule 2.

    Model B is a very safe model to use. No problems with Kelly. The drawback is, that with level staking most bets will be underbet, some very much, and that will hurt the profit.

    Model C is the theoretically best model. But we can’t use it because it requires variable staking and information on the edge for every bet.

    My ambition was to find a hybrid model between model A and model B. Unfortunately I have not done any progress despite hard work.

    Personally I will continue to use my own variant of an approximate Kelly model.
    I feel comfortable with it.
    In last season it gave me a ROC of 91% and this season stands at a ROC of 120%

    I hope you have enjoyed reading some of my thoughts on betting. I know that not everybody see it the same way as I do. But is it not the beauty of living in a free part of the world where we can argue, disagree and learn along the way.

    Thank you and bye.