Saturday, 15 May 2010


And time to breathe even, let alone make a decent post on the blog!

Project alpha which has been undertaken this week is going very nicely - looking at when to place bets, and in more detail about how and why prices move. Goes without saying that if you can beat the closing lines at oddsportal (which I was doing with regularity) you have a very good chance of beating the game even if your picks aren't that great (which I hope mine are of course!)

Project beta, interpolating data about minor leagues around the world and combining some ratings to find opportunities around the world all around the year, be it in japanese or hungarian leagues (!) or some even more bizarre than that, is only 2 days in (and quite labour intensive at the moment, sadly, although that's something that could be worked with).

With optimized stakes (or something like that anyway, currently using 1/10th kelly since it seems to fit the profile reasonably well, personally i hate getting bogged down with staking), the overall picture is losing 7% ROI (betting on a high proportion of games). At the mo I haven't filtered any out (apart from a couple of italian games where the draw is 1/2, 4/6, prices like that) for being at the end of the season, as I would do normally. And its only been 2 days! However during my testing and sewing together phase I noticed that the model is overkeen to oppose very short priced teams, and I find this to be a common problem with fairly simplistic models.

By this I mean that its very rare that you see many people (who use modelling) offer up Chelsea as a value bet, but that's not to say they can't be (i've been confident each time they've been highlighted on the blog, and I think they've won every time from memory). It's more "fashionable" to be taking such teams on, and away from home I'd support that on a fairly regular basis - but at home you can see by looking at the markets for the last few years that things are evolving. It used to be the case that a standard home handicap for Man U was 1.5 asian goals, now its more like 2 or even 2.25 against the lesser teams. I actually think it takes quite a bit of courage to back the shorter priced teams, since apart from anything else, lumping on short prices willy nilly is the best way to do your coconuts!

So I decided to filter out any games where the underdog is more than a 0.75 goal asian handicap underdog (their price is over 2.00 + 0.75 AH). This keeps the no-hoper bets out. You can't gain much from backing some of these teams at 27.6 and prices like that - until you have 10000 games worth of data (IMO). You can draw much more solid conclusions from a more compact dataset of prices from 1.4 to 6.00 I would say (arbitrary values). Also however if the model DOES highlight a short one as a good price, you should probably listen, since its almost inevitably good value.

From that filter I was only left with 9 bets in the last 2 days, yielding 19.77%. Tiny sample and very early days - but what I did find encouraging is that some of the bets were subject of severe market support (which is normally a very good indicator that you are on the right side of the bet, in my experience). At level stakes the yield was down to 2.44% which might show that the optimized stakes are working (although over 9 bets, there's just no point drawing any conclusions!).

Overall I'd expect the optimized stakes to make 1-2% difference to long-term yield. If they look like long-term profitable bets, with a 10%+ yield, when I have a sample of over 200 games, I'll start posting them on the blog - with the rider that I fully appreciate some of them will be difficult or nearly impossible to get money matched on, because they will be bizarre leagues and only a handful of bookies will offer the prices (which in itself is quite heartening, since you'd expect the opportunities in such leagues for massive pricing errors to be vastly more prevalent than they are in the premiership/la liga/serie A).

Keep your eyes peeled!

1 comment:

  1. I hope to have the time at some point to go back through this season's bets where possible and back-test how they would have been rated under these new conditions. Should be interesting!