Tuesday 25 February 2014

A quick look at value ratings this season for unique bets

(The below was written before tonight's bets. I meant to post it last night but didn't have time)

Before I get onto the purpose of this post, I want to say a few words on the discussion I had with Tage in the comments section on the last post. Tage has been a member of TFA for a while (3 seasons I think) and therefore, we’ve exchanged numerous emails on the performance of the systems and the ratings and Tage has always been supportive of TFA. Over the years, we’ve not always seen eye to eye on everything and I know Tage has his own views on the service, the systems and what I should do differently in terms of following the systems and running the service. However, like all TFA members, I respect his views and I agree with a lot of what he says but I also have my own views (as do others) and therefore, I do what I want to do with the ratings, the systems and the service.

Tage was a little annoyed that I butted into his conversation on the last post when he said Vidar’s staking plan was “wrong”. There is no right or wrong way to follow TFA and therefore, I’m not going to have one member telling other members they are doing things wrongly. Everyone can do what they want with the bets and if someone wants to have a staking plan of 0-1,000,000pts, they can do so as it’s their choice. Might not be the staking plan Tage would choose but it’s not his staking plan. 

I think we’re all still learning about TFA and the ratings, the systems, the draws, the Euros and so on and I expect it will always be a continual learning process for all of us but that’s part of the fun I think of following TFA. It’s what makes TFA fairly unique and ultimately, it’s why some people follow TFA believe it or not. It’s a challenge at times but I’ve had so many notes over the years from people talking about how good TFA has been for their betting, not only with the profits from the systems but the way it allows you to do different things with staking and portfolio bets and it has taught them a great deal about betting and how to make a profit and therefore, I don’t want that element curtailed.

I think it’s great Tage shared his thoughts on the blog and as subscribers will know, I allowed Tage to share a report earlier this season on why staking so much on some teams wasn’t worth the returns for the additional volatility it brings and I think all of us can learn from Tage after what has happened this season! However, at the end of the day, Tage can’t be telling people that their strategy is wrong when they are in profit and others are not in profit!

Anyway, I apologised to Tage for stepping into the discussion but I don’t want people thinking they are doing anything wrong when they are just doing their own thing with the staking. Tage laughs at this as he’s commented before on the blog but I know one guy who follows all (and I mean all!) TFA bets every season. He’s not had a losing season so far, is on course for another good season and constantly tells me that everyone else is crazy for not following all my bets! If anything ever points to TFA being unique, that is probably it. :)

Right, the purpose of this post is to answer a question Matthew raised on a post a few weeks back. I said I would do it at month end but I’ve had a little time today and it’s a fairly quiet midweek with fixtures, so I’ll do the post now.

Here’s the question from Matthew again:

Hi Graeme,

Could you provide an update of the results of backing each individual selection once to 1pt for the season so far? If you could then split it into value bands as well that would be great.

Regards,

Matthew.

I’ve pulled together the analysis for the UK bets and I think it’s very interesting. I keep saying it but with the introduction of me releasing the value ratings this season with the bets, it does add a new dimension to the systems. Admittedly, I chose to ignore this dimension this season even though all the evidence pointed to possibly removing the lowest value bets from the systems historically. Having done this analysis today on this season, I suspect the evidence is even more overwhelming now but like always, I’ll reassess things at season end and give my view. I don’t fancy removing the low value bets completely as it would lead to a big reduction in bet number but if the evidence is there, the least I can do is point it out and people can make their own decisions as always.

Here’s the results this season for the 4 rating algorithms (data to today), split by value band.


This shows it was a good weekend as remarkably, all 4 algorithms have now sneaked into profit. As usual, algorithm 3 (system 31) is wiping the floor with the other 3 algorithms. I guess the surprise is that algorithm 1 is beating algorithm 2 and historically, this definitely hasn’t been the case. System 21 is a better system than system 6 IMO and the results back this up.

As you will see, the results by value band are interesting. Systems 21 and 41 show large losses for the lower value bets (<10%). System 6 shows a large loss for the lowest value bets (<5%) but a great profit for the 5%-10% range and then a large loss for 10%-15%! System 31 bucks the trend and the low value bets do really well.

Looking at the other end of the scale, the highest value bets (>30%+) are having an amazing season on every algorithm! Admittedly, there are so few bets, I won’t be retiring just yet but it at least points to the fact that if you are only wanting to lump on a few teams a season, these are the teams to do it! :)  (That’s tongue in cheek)

Being honest, as I keep saying, the only conclusion we can draw is that generally, the higher value bets do well and the very weak bets seem to be doing badly this season.

The actual question Matthew asked was around unique bets and not the algorithms, so here are the results for the unique bets at an overall level:


First comment is that it’s not exactly the best return in the world if you are following this strategy! You’d be much better just following system 31 as this return over the seasons has easily beat this strategy of backing each team that appears once.

Of course, to break the unique bets down by value bands, it’s not easy to just have one set of value bands. If we take algorithm 1, not every unique bet appears on this algorithm and therefore, we’ll have a bucket set of bets that don’t appear on the algorithm and therefore, bear this in mind! It starts to get quite complicated when you think it through as system 31 is the best system but this system seems to find bets that other systems can’t find and therefore, this should play through in the results here (and it does!).

Right, here are the results for each algorithm for the unique bets only:


I’m not about to sit and spell out hundreds of observations but I think system 31 is interesting. If you take the 212 bets that don’t appear on system 31 but appear on one of the other 3 algorithms, this accounts for a loss of 44.5pts from 212 bets. Not a huge sample but given system 31 is the best system historically, I expect that if system 31 doesn’t like your bet, not sure it’s worth having!

That’s a pretty good place to be and at least gives us some food for thought I think about the future. Likewise, if your unique bet only has a value of less than 5% and appears on system 6,21 or 41, it’s probably not worth having either!

You start to put together a couple of rules to build yourself a little unique bet system and quite quickly, you have a system that’s much better than just backing all unique bets. 

As always, I prefer to draw conclusions at season end and on much bigger sample sizes than one season, so please don’t start doing things off the back of this post. The results this season with the value bands fit in with what I thought would happen having spent some time with the data during the Summer. That's pleasing although it's one step further to act on this. 

For those reading this and thinking, ffs, he’s making it even MORE complicated, I will try to improve the service next season by having some systems that are based on certain rules regarding systems, ratings and draw systems and ultimately, these may be useful to some people. Not sure how they fit in the service and maybe they sit in another service where people are told what to back rather than told to do what they want to do! 

As always, any questions and observations welcome!  

5 comments:

  1. Hi Graeme,

    Thanks very much for your detailed response it is much appreciated. I like to see the unique bet figures as it represents the most diversified, lowest geared portfolio. I would be interested to go one step further and see how the unique bets have performed with double staking the homes as I believe that many people employ that method. The fact that the unique bets generate a profit would suggest and sound foundation for further filtering.

    The value rating analysis I find very interesting. My experience historically is with statistical based football models (non player based) which will provide me with a goal supremacy and total goals calculation which in turn are plugged in to a relevant distribution for that particular league (not all leagues have the same goal distribution). In this type of model I have found that as value increases so does profitability but only to a point. I look at value in absolute terms rather than relative so if I make an event 33% likely and the odds suggest 30% probability I think of this as 3% value rather than 10% value. I find this clearer as it means that big odds selections don't appear to have 50% value say which feels like a misleading number in terms of representing bet merit which of course it isn't but there you go. So I find that up to about 10% absolute difference there is a consistent increase between value rating and profitability however, beyond that there is actually a decrease. This would suggest that at a certain point the model is just missing a vital piece of information that the market has accounted for (missing key players etc). I don't think I have had a 30+% value bet ever in my life with this style of modelling. I find that 3% - 10% is the real sweet spot for me.

    Regards,

    Matthew Trenhaile.

    ReplyDelete
  2. Graemes fundamental thinking about football betting is centred around the idea of SELECTION and SYSTEMS as far as I can tell.
    There is nothing WRONG in this approach.
    And it is shared by Vidar, who writes:
    But I can´t understand your thinking. Staking 4 stakes on each game? I can´t just ignore my systems. If a game appears on all 15 systems, then I have to play it, staking 1 point on each system (=15 points stake in total). Otherwise, what´s the point choosing the systems if I am not going to follow them?
    There you have it. My proposal was to use another STAKING method.
    In this blog there have been written very little about STAKING, but an awful lot about systems and portfolios.
    And my observation after reading this blog post is, that Graeme

    will try to improve the service next season by having some systems that are based on certain rules regarding systems, ratings and draw systems

    So more systems, not a word about STAKING.
    I was one of those members that urged Graeme to include value ratings.
    Because if used wisely it’s possible to make more profit using a BETTER staking strategy.
    I’m still not saying that the method TFA uses is WRONG, I say there are BETTER methods.

    Go to this site: http://www.sports-tipsters.co.uk/leaguem.php
    Number two on the list, Sportyy, is one of only two here that has shown statistically significant results.
    The service is Spanish and they use fractional Kelly staking, a variant of the method, I described in my last comment in the Vidar blog.
    Steve is Australian and most of the models, he follows, are from an Australian service called Sportpunter. Sportpunter uses fractional Kelly staking as well.
    Just take a look at the results Steve has obtained via these models.

    Kelly staking is based on value ratings like the TFA value ratings.
    In my opinion the next step forward for TFA is to seriously think about how these ratings can be used in offering alternative STAKING methods and not only more SYSTEMS.

    ReplyDelete
  3. Hi Matthew.

    I agree with what you are saying completely. There are numerous examples thrown up every week with my ratings (mainly in the Premiership or SPL as I watch this footie more than the lower leagues!) whereby I know the ratings are missing vital pieces of information and the value % thrown up are meaningless but I’ve no real desire or need to develop my ratings further. The best example was Newcastle the other week (think it was away to Chelsea) when my ratings were massively out of line with the market. Someone asked me if the ratings had an issue but the issue was missing data. In this case, it was a key player who had been sold (Cabaye) and a few injuries and therefore, Newcastle had drifted wildly by about 30%+ if I remember before I collected the odds and therefore, the market had adjusted and my ratings obviously can’t adjust.

    I’ve made it clear before that the value ratings aren’t the be all and end all and I don’t support anyone using Kelly staking or any variable staking plan on the value ratings as the value ratings are only one piece of the jigsaw for my systems. The value figures are only an indication and I don’t really believe I’m finding 30%+ value in games although so far, the results indicate I’m finding more value than this although it’s just variance due to small sample sizes.

    My hope is that generally, the higher the value, the better returns but as I’ve said before on the blog, AT THE MOMENT, I personally don’t care where the profits originate from. My own profits depend on the system bets I follow!

    Going forward, I might care a little more if I start using the value ratings as a filter and maybe adjust my staking plan but I’ll see what I think this Summer when I have more time to analyse things.

    Cheers for the comment and glad you liked the analysis.

    Graeme

    ReplyDelete
  4. Hi Tage.

    My fundamental thinking about football betting is pretty simple, I want to make as much money as possible each season. I want to take lots of risk and I want to make lots of money. I aim for 100%+ ROC each season. To achieve this sort of return, you need to be taking some risks and so far, I’ve done OK.

    In addition, if I go broke, I really don’t mind (OK, I do mind but I try to put on a brave face!) but this isn’t my job. If anyone wants to pay the prices the other services charge, I think the least they should get is a good staking plan and it’s good to see that this is the case. For the prices I charge and the results, I think you’re asking a helluva lot for me to provide a staking plan!

    I’m sure the other models you refer to use player based models and are much, much more accurate than my ratings. My ratings are pretty basic (as I freely admit) and they give me an edge betting early into the market but that’s all I’m providing.

    If I really wanted to do this properly, I would provide you with a list of all UK and Euro matches every weekend, all 16 ratings for each game, all historical results back to 2000 with the 16 ratings and you could all spend days, weeks, months and possibly years developing systems and staking plans but I’m not providing you with this. If I get to where I want to with my betting, then I’m not exactly going to share with you everything that got me there!

    I’ve never hid the fact that the betting side of this is more important to me than the tipping side. The fact I give away the bets so cheaply is evidence of this along with the fact I cap membership. The reason the odds move so quickly for Sportyy and Sportpunter is because they have an edge but they rake in money from subscription fees. They’ll make 3-4 times what I make a year betting and through subs Tage.

    I don’t think TFA will ever be the service you want it to be as I really don’t have the inclination to go down this route. It isn’t aimed at mathematicians and statisticians (I seem to attract plenty of these types though!) and therefore, there may be better services out there.

    I know the frustrating thing for you is that you see the edge I have (even if it is not statistically significant!) and you think I can do more but I enjoy what I do with the footie Tage. It’s pretty simple, I enjoy following the bets, I enjoy my staking plan and I enjoy TFA. The day I stop enjoying this, there won’t be any service!

    So, in summary, it’s not my job to supply you or anyone in the service with a staking plan. Come up with your own staking plan. ;)

    For what I provide, at the price and the fact I quote realistic odds that means subscribers can get my prices and make a profit, I think I do OK.

    Graeme

    ReplyDelete
  5. Hi,

    Of course the most optimal staking plan is Kelly criterion. But it has at least two big issues:

    - The value is unknown. Truly, with each TFA system bet we get the percentage deviation between odds and probability estimated by system. But, as Graeme reckons, that percentage is useful to determine the most valuable bets, but the exact value is not "real".

    When I want to apply Kelly criterion, instead of using value, I prefer to use historical yield to estimate value. In this case, it's easy to see that stake=historical yield/(odds-1)

    - Kelly criterion penalizes bigger odds, which are often those that have the biggest value

    Diego T.

    ReplyDelete