2012 and beyond

Results are continually mediocre, and it seems likely to me that a relatively "scattergun" approach which worked for me up until 2011 started, is having its problems.

I will list a few that I think there are:

i) Football ratings are now the "trendy" way to bet on football. Blogs pop up all the time, the models are not incredibly complex (although everyone has their nuances), extremely competent ones are even available for free online if you know where to look. They are all over SBC. There is also a tendency for the very best stuff not to make it out into the open market, of course.

ii) Books are getting smarter. They often delay pricing games until the first asian lines have come out. In the recent years the asian markets have got better and better (as one would expect) and those outlying 6/5 cases are becoming 11/10 and evens more often - the premiership bears this out more and more. Teams that used to be 6/5 regularly are much more often 2.04. Just as a few years ago the likes of Man Utd were at LEAST as superior c.f. their opponents as they are now, but you would take -1.5 to back them on the AH, not 1.75, 2 or more.

iii) The successful modellers get bigger and bigger bankrolls, and start betting earlier and earlier. I know a chap who bets on sunday, as soon as he can download the stats. Its all automated and I personally can't compete on that level. Even having said that, like almost all ratings players, he is struggling this season and is still making a net loss.

iv) services are now looking for other reasons for their poor(er) performances compared to the past, and are looking to cut each others throats on release times etc. That will inevitably end in tears, far more money pumped in by the sheep in the very early sensitive markets, most people will not get the prices released and rather than 10-40 people getting on early at the "proofed odds", it will be more like 2-8. This is why I have always proofed at closing lines, because they are the only fair gauge of what is really widely available. If Coral are 4/1 and the next best is 4.75, I don't have too much of a problem with it (although I will never get the 4/1). If they are 4/1 and the next is 10/3, and the average price is 10/3, then the problems will start, and claiming a 4/1 winner is only cheating the record (IMO). I respect the tipsters who proof at average prices on oddsportal or similar, or generally available prices, because they are actually achieveable for

Looking at my results for the season, and in the past, I am seeing a much better set of results from the best/strongest games. This has been borne out this year in the performance of the 1.5point and above bets. This is not a recurring pattern over the years however, which strongly suggests that this is variance. Its not news to me, and indeed backs up my own feelings about using Kelly, etc.

However, the strong bets have performed most consistently. I seek to provide a minimum of 5% ROI, otherwise I see no point sharing the bets with the blogosphere in the hopes that we can all make some profit. This seems to be best served by concentrating on the strongest bets, and making them yet stronger. I don't want to cut down to 75-100 bets per season particularly, because there is so much potential for variance to really rear its head, which is why i have always stuck with many more plays than some other sites.

There will always be a "backtesting" bias in the forms of modelling used. I have never seen the results from past histories replicated, which is why I strongly dislike a lot of the stuff that certain websites and forums churn out, especially when they don't make it 100% clear that "results" posted on a site are NOT results in action, but results of a model run over "new historical data". Its very misleading at least, if not wholeheartedly dishonest. I also would take any results not achieved within the last 24 months with a pinch of salt - the marketplace is dynamic, and it was very very much easier to make money in 2008 and before (from my own personal experience). Things move along quickly for the reasons above and other reasons. I would discount anything from 1-2 years ago quite keenly as well. Variance that is being exhibited is either 2+ Standard Devs away from the mean, or else the underlying assumptions are flawed. I favour the latter explanation, because this is not borne out in other forms of sports modelling in which I am involved. Yes, home and away wins can vary 6-8% per season, but that's nothing different from a number of other sports and areas that I have experienced.

It also makes more "sense" that there will be a handful of stronger plays rather than 10-15 glaring errors per matchday. The books do not give their money away. The strongest models use underlyings as only 30-40% of what they do, and there is a reason why these guys have built companies employing numerous quants - they are very very good at it. They are also using stats well beyond the capability of football-data etc.

So, I am going to look to become a LOT more stringent on play selection. Expect less plays of higher quality, and probably all of one unit only. I may well continue to track some plays, although I struggle to say the least with the results side of things, so I will have a think about this.

Here's to 2012 and putting on some points for the rest of this season, which has been tough so far for most in the soccer modelling game.