I’ve not hidden the fact that one reason I believe my systems are much better than everything else I’ve come across is because I am working with more than one ratings algorithm. Hence, when I combine rating algorithms, it ensures that only the very best bets make it onto the combined systems 6-21 etc.
Of course, this season, I tried to cross refer my new 3rd rating algorithm with 6 and 21 to try to see if that works and being honest, it has done OK. I’m not sure what I expected but if I look at the returns for each algorithm individually and then the combined systems:
6 - 6.1% ROI
21 – 9.4% ROI
31 – 5.6% ROI
6-21 – 11.0% ROI
21-31 – 8.7% ROI
6-21-31 – 9.8% ROI
Clearly, cross referring the algorithms tends to improve the ROI a fair bit this season and backs up everything I’ve shown with the backtested results and also of course, last season when the systems went live. Interestingly, 21 does better on its own rather than cross referring with 31 this season but I suspect it’s just volatility. Historically, this hasn’t been the case.
Therefore, we know that cross referring the bets on different algorithms improves the returns.
I guess one step beyond this is actually trying to analyse the returns based on how many times a team appears on each system. I have 20 systems this season and therefore, based on the backtested and live results, how have these teams performed depending on how many systems pick them?
I did this analysis a few months ago in December ( see this post http://the-football-analyst.blogspot.co.uk/2011/12/i-thought-id-post-this-piece-of.html ), so the results won’t be too surprising for long-term readers but one thing I couldn’t do then was show the results split by Home/Away bets. I’ll cover that in this post though.
So, firstly, here are the returns for all games based on the number of times each team appears:
I realised people may struggle to get this concept but quite simply, it is the return from staking 1pt on a team that appears 20 times, 19 times and so on. Hence, there is no direct link between these results and my overall proofed results. My proofed results are 1pt win on every team on every system, so 20pts on teams who appear 20 times, 19pts on teams who appear 19 times.
These results are much in line with what I posted before. Since the last post, there appears to have been 10 teams that have appeared who have appeared on all 20 systems. The profit from backing these 10 is 7.6pts. Hence, a 76% ROI!
Anyway, I don’t want to dwell on this again as quite simply, the graph of the cumulative ROI hasn’t changed too much. You will see I’ve added AH +0.25 and AH +0.50 to the results since then as there are two extra lines on the chart!
So, that’s an update of what I showed before. Now onto more interesting things. The split between Home and Aways. I said this caveat on the original post on this subject but this type of analysis I’m doing isn’t easy. It’s not that the concept is difficult, it’s the fact there is massive data manipulation needed. I noticed that summing the Homes and Aways doesn’t quite get you back to the overall figures.
The reason being that any time there is a spelling mistake for a team or a wrong date against a game, it won’t recognise the fact it’s the same bet on another system. Unfortunately, when you are recording as many results as I am and sometimes websites for fixture lists change the names of teams, then it means that I can quite easily record a team like Newport as Newport, Newport Co or Newport County. I try my best to look out for this sort of thing but I sometimes have to manually update results when data isn’t available and that means I can sometimes change the name of teams on different systems or record different dates against teams on different systems! Overall though, nothing like this can invalidate the results or the conclusions but for someone who’s as anal about data as I am, it is annoying when I notice these things!
Understanding the above then, let’s now look at the results for Home Bets only.
One thing I find with this sort of analysis is that there is too much to look at and take in. In addition, what I look for may not be what someone else may look for, so I’ll keep this high level but you’ll get more from this by analysing the table yourself.
There are two things I’ll pick out. Firstly, there have been 78 bets overall that have appeared on 20 systems. 59 of these are Home bets which is a little surprising but then again, I know why this. The strongest bets my systems find are actually Home bets. Hence, it is easier for all the systems to agree on a Home bet rather than an away bet. Being honest, if system 22 likes a Home bet, there is a strong chance every system will it as 22 is the strongest system for Home bets.
The other thing is actually more important to me and that’s the fact that there is little difference in return from backing Homes when they appear on anything from 14-20 systems. Hence, if you were wanting to follow the Homes using this sort of strategy, you could do well just following the teams that appear on at least 14 systems. 323 bets to date and 103.5pts profit. Not bad at all!
Interestingly, Home bets that appear on 10 or less systems have only made 170pts from 1,889 games. Not great but then again, it’s still a 9% ROI on Home bets which are lower odds, so the ROC etc. will still be decent enough I suspect.
You can also see the big difference in the returns if you try to cover the draw in some way. As I’ve said numerous times already, you can’t get the draw on your side on Home bets as it is giving away nearly all of the edge. Saying that, if you were backing Home teams that appeared on 10+ systems, you could still achieve a 10% ROI by using AH +0.5, so it’s not that it isn’t possible to make money this way, it’s just not the optimal way based on ROI. I can’t even guess what the ROC would be like on this as I can’t imagine too many big drawdowns backing teams to win and draw at home!
OK, let’s now look at the Aways. Here’s the results:
Firstly, you can see the shape of the graph is much different. As you reduce the number of times an Away bet appears, the returns drop quite significantly. Hence, it is actually the Away bets that give the overall graph it’s shape if you know what I mean.
You can also see there are a lot less Away bets that appear on many systems but in general, there are a lot more Away bets! Again, this isn’t surprising as it’s pretty easy to find value for Away bets as they are the place where most punters overlook but as I’ve said before, you don’t want to be lumping on big value Aways with high stakes due to the volatility aspect.
You can see that covering the draw works much better with Aways too as there isn’t as big a gap between the H/A returns and the other returns.
I think that will do for now. You can see the power of this sort of analysis though but being honest, I could write pages and pages on this and how it could be used to help determine an optimal staking plan for each follower who wanted to follow this sort of strategy but there are only so many hours in a day!
As always, any comments or questions, feel free to leave a comment and I’ll get back to you when time allows.
great post, whats "ROI"?
ReplyDeleteHi Al.
ReplyDeleteROI is simply Return on Investment. It’s the return divided by the total staked on the system.
ROC is often referred to by me and this is Return on Capital. This is the return divided by the betting bank.
I discussed this before on the blog but in my opinion, ROI for show, ROC for dough. What tends to happen is people find systems with massive ROI’s but the betting bank needed to follow is large also, so when you look at ROC, it isn’t that great.
The secret is to find systems with a high ROI but more importantly, if this is done following fairly low average odds, then the ROC can be very large as the betting bank needed will be fairly small.
In the first season, I tripled my betting bank before losing 1/3 of it towards the end of the season (I doubled my bank that season) and this season, I’ve doubled my bank again. However, it’s all relative though as it depends on which systems people play and what bank they set.
As I’ve discussed before, ROC is such a subjective measure (people who take more risks get a higher ROC) but when comparing systems from the same set of systems (i.e. mines), it is OK to compare ROC as they are comparable. What you can’t do is compare the ROC on my systems with other external systems unless you adopt the same criteria for setting the betting bank.
Hope this makes sense. I think I accidently deleted another comment of yours last week asking for info on how to join the service but I’d suggest you just keep an eye on the blog. Unlike last Summer, blog readers will be made aware of the opportunity to join the select service.
Cheers,
Graeme