Wednesday 13 June 2012

And now the second European algorithm (System E6)


There was a good question left by The Portfolio Investor (Rowan) on my last post who commented:

OK, I'm going to be getting a little ahead of myself and yourself here, but there has been a recurring theme throughout these posts on the Euro systems. I've a feeling you may well be intending to address this point a little further down the line. If so, tell me to pipe down and be patient. :)

Being straight with you, my intention has always been to follow the Euro systems next season, but not with hard cash. As you say yourself, next season is to be judged on the performance of the UK systems and that's good enough for me.

But, the recurring theme that you keep touching on are the similarities to the results that you experienced when you were developing the UK systems. I see there are differences between weightings of home bets vs away bets, but in terms of the results..."Quite spooky how close these returns are to the UK systems!"

Should these similarities not give us a lot more confidence in backing these systems? Am I being a little too cautious in thinking I should just watch these systems perform over 2012/13 to see how it goes? I'm not suggesting we wade right in (you know me! :) ) but at the same time...

Right, tell me to bugger off if you like, and to stop trying to rush things.

Rowan

I would post up my reply back but it’s about as long as my blog post below and all I said was that I’d do a blog post on the topic in a few days! Basically, it’s a good question as it’s one I had going through my head anyway.  What should TFA followers do with these new European systems next season? 

I’ve said to Rowan that I’d try to weigh up the pros and cons of deciding whether or not to follow the Euro systems next season and I’ll cover it in a blog post. It’s an interesting one as it’s easy enough for me to say that we should all just ignore the Euro systems next season and watch how they do but do we really learn much from doing this?  I don’t know and as I said in my reply to Rowan, I’ll be following some of the systems myself anyway as it’s what I do with new systems but I’ll stake them properly to ensure that any damage will be limited (of course, as will any upside!)

Anyway, save that for a few days time. More pressing matters ahead.

We’re now onto the second European algorithm I’ve created. In essence, the next two systems are the brothers of UK systems 21 and 22. 

This is the first look at system E6.  Given the comments on the last post about systems 6-8, it’s probably worth knowing that system 21 had an ROI of 23.1% during backtesting. 

OK, so here’s the first look at the results of system E6 by season:


Right, that’s a very different looking table from system E1 and the results by season are intriguing.  The seasons used in the backfitting process are identical as on the first algorithm, so the returns in seasons 2006/7, 2007/8 and 2010/11 may be a little overstated.

Looking at it overall though, an ROI of 17.2% from 4,475 bets. It’s actually very close to system E1 if I’m honest but it is definitely not as good as system 21 in the UK systems.

I think you can definitely see the impact that the backfitting has on 2010/11 but for some reason, it doesn’t show through in the first two seasons.  Without getting into the real technical aspects of why this may be, looking at 2010/11, I’d say that the results are a bit too good if there is such a thing.  Likewise, I would say that the results in 2006/7 look too bad to be true! In all my system building, season 2006/7 has always had the best results until this algorithm has appeared now.  Not surprising really since 2006/7 season is closest to the earlier seasons which are used fully in the backfitted process.  Hence, if using data from 2000 to 2006, you would expect this to be the best indicator for season 2006/7 rather than any other season going forward.

What does all this mean? Well, I don’t really know. I hadn’t expected to see this, so not sure what I do. I’ve checked back at the work and I can’t see anything that’s wrong. The seasons all look correct.  Seasons 2000 to 2006 all have ROI’s in excess of 20% and are fairly stable season on season. However, there is a drop off in 2006/7 to 2008/9 and then the algorithm picks up again from the 2009/10 season.  Blips in 2010/11 which is not unexpected and looks fine last season.

I think until I get into looking at the analysis of the results for this system, it’s hard to pinpoint if there is a fundamental issue with the second algorithm or whether it’s just a funny within the data for these seasons.

One reason for having more than one algorithm is for this very reason I think. I’d be slightly worried the second algorithm doesn’t show the same traits of every other algorithm I’ve built to date but then again, it’s only system E6.  How do I reduce this risk? Well, I could cross refer this system with E1 and check when they agree on bets…..Ah yeah, create combined systems, that’s a good idea to reduce the risk. Who would have ever thought of that? ;)

Anyway, joking aside, I think this shows why the combined systems are important.  Yes, if I was placing my life on system E6, I’d maybe be worried something has gone wrong within the system build but if I stick to betting on teams when both algorithms agree, it will surely help minimise the risk a little!

Interestingly, due to the fact the partially backfitted season’s don’t all look great, here’s the results if we exclude these seasons:


Now, this is a first whereby the seasons with absolutely no backfitting involved experience nearly identical results to the seasons with backfitting involved. Is this so bad?  Well, I’d like to think it maybe turns the above musings on their head and says that system E6 is maybe a better system than any other system I’ve ever built?  Of course, I don’t know the answers to most questions I’m posing (I never will!) but it’s important to always be asking yourself these things and challenging your thinking. I give myself a hard time all the time about the questions I'd want answered!

Right, so how does the split between Homes and Aways look?  Here it is:


65% are Home bets which is slightly more than on algorithm one.  An ROI of 14.4% on Homes and 22.6% on Aways.  Very same comments apply as I’ve made before on E1-E3. I would have expected a higher ROI on Homes and a lower ROI on Aways.

Lastly, here’s the split by League:


This is a bit different than algorithm one.  Spain has the highest return from Germany. Again though, the German league has a few number of bets than the other leagues.  Being honest, I hadn’t picked up this theme when I was building the ratings, so there is clearly something that stops the German league from having more bets.  It’s easy now I’ve written it like that isn’t it?  Yeah, there are only 18 teams in the German League! Hence, 4 games less per year than the other leagues.  (Another example of me being a fool!)

So, an interesting introduction to algorithm two for the European Leagues.  Doesn’t quite live up to system 21 in the UK leagues but then again, not many systems do!

Right, I'll be back with system E7 soon and then we’ll have a bumper post on the combined systems I think.  Going to do all the 6 combined systems at once in a single post (allows me to get on with producing the spreadsheet for download with the analysis of all systems) and it gives you guys a wee while to digest the introduction posts for all the European systems.

Any questions so far?

No comments:

Post a Comment