Similar Games Model

I'm not sure if I'm suffering from some sort of addiction to Excel but you won't be too surprised to hear that I'm looking at potential new football systems for next season!

I think my one worry with everything I've done on the football modelling so far is that it is all based around a ratings approach to football modelling. I won't try to pretend to anyone that I believe there are better ways to do this as this approach is one that has stood the test of time and ratings are used by many people to correctly predict football games. I'm also using factors that are widely believed to be the most successful factors when building football ratings, so I feel like I have most angles covered.

However, what happens if my approach suddenly stops working as well as it did during backtesting and during last season? This could be caused by inherent flaws in my models that haven't come to the fore as yet. It could also be caused by others cottoning on to what I'm doing and building similar or better models than I'm using. Lastly, but the least likely, it could be caused by bookmakers changing their odds in such a way, that my edge is eroded. I don't think my 9% return last season will have bookmakers giving up just yet!

There are clear advantages and disadvantages to using ratings for football modelling but over the last week or so, I've started to think that I could use my data to look at another ways to build systems.

Now, before I start on this road, I don't want anyone to think I'm going to end up with another 10+ systems and I don't want people to get too lost in what these new systems say.

To put it in perspective, if my new systems (31-33) carry huge risk next season, then any other systems I'm tracking for next season are going to be off the scale when it comes to how risky they are. In addition, they won't be based on my good old trusty method of ratings analysis, so they might turn out to be rubbish. Hence, these really are paper tracking systems.

Interestingly, I was exchanging emails with a blog reader yesterday and I happened to mention that I was thinking of doing something outside of my current ratings methodology and once I explained what I was thinking of doing, he wrote back to me and said there is actually a name that is used to describe this......

'Similar Games Model' - I learn something new everyday!

Basically, my idea was to trawl through my data and try to build an algorithm that identifies historical games that featured statistically similar teams and try to see if I could find an edge with these games.

For example, I find a team who's coming off 6 defeats, playing a team who's won their last 6 games. The form team is a heavy favourite and the other team a big outsider. Is there an edge here of backing the underdog or even the favourite?

Now, that's an extreme example but you get the idea. I write something that can pull out games where the teams involved are statistically linked in some way. I then see if there is a way or determining whether or not there is an edge there. As a last check, I'll apply my ratings from the 3 algorithms I've built and see if any edges appear that I can maybe exploit.

Well, going by the fact I'm writing this post, you'll know that I've got something already but unfortunately, it was in an Excel sheet and I should have put more thought into the output format! I thought putting the output into different Excel tabs would be a good idea but I've used the maximum memory I think my PC has and ended up with a lot of tabs!

After a bit of manipulation (thank God for the Indirect function!), I've got some data out that looks sensible.

As you can imagine, there are lots and lots and lots of ideas here but they fall over due to lack of volume of bets. I'm not in the game of adding up different ideas to make a system and therefore, you either find something that works for a fairly large sample or you don't find something at all.

Obviously, you can relax the criteria a little and widen the rules and that creates enough of a sample although it dilutes the returns a little possibly as if you have a criteria that works, expanding it will lead you to include more rubbish!

Anyway, the crux of it is that I have a selection of systems that throw up bets. I'm at the stage now of going through each system and checking what sort of system it is and whether I could use it next season.

I think I'd ideally be looking for 3 systems max to follow as it would be interesting to test whether this methodology works.

For me personally, it would be really interesting to see how these systems would compare against my other systems.

Watch this space......