I’ve decided to take the easy way out with the Under/Over 2.5 goals systems and ultimately, I’m going to have a second shot next season with the same systems as last season. I did intend to rebuild these systems after their poor performance last season but when I look at the results, maybe it wasn’t quite as bad as I thought last season and therefore, I’m going to apply a filter on one of the systems, keep the other system the same, drop the systems with not enough bets and we’re set to go for next season now. This also ensures that the proofed results from last season are also fairly accurate as I have always said that even if I rebuilt the systems, they would keep the proofed results from last season, regardless of how I chose to improve the systems for next season.
As I’m aware, this is actually the first time I’ve discussed these systems on the blog. Someone asked me very early on last season if I had ever looked at developing a system for this particular market and I gave a really cheeky reply back that I thought anyone doing this was wasting their time. I don’t know why but I’ve never really seen Under/Over goals as anything too different from someone betting on the number of corners in a game or the team to get the first throw in. Admittedly, this is a very crude way of looking at it and I know there are many punters/traders who use the 2.5 goals market effectively and manage to turn a profit over time
Curiosity got the better of me and we then got into a discussion about how you’d find an edge at this game. Quite quickly, I realised that the conventional thinking of looking at recent games, working out how many times games involving the teams had been over or under 2.5 goals was flawed in the sense that where is the edge from doing this? If everyone can see that two low scoring teams are playing one another and the game is going to be tight, then how do you find an edge in this market?
Similar to my ratings for calculating value in match bets, the answer has to be to find an underlying indicator that others can’t use. By others, I am referring to other punters mainly but of course, I am also referring to the odds compilers as at the end of the day, my system only has to be smarter than the bookies setting the odds. Once I’m on the bets and everyone else then comes around to my way of thinking, it makes no difference to me that the team shortens or drifts as I’m already on the bet.
Clearly, I have a set of ratings which work well for match bets and all I had to do was to try to find a way to get the ratings to find value bets in the Under/Over 2.5 goals market. It was only about a week’s work I think to gather the data, build a value model and develop some systems. I then applied the same methodology as I do with all systems and ended up with 3 systems for Under 2.5 goals and 3 systems for Over 2.5 goals. The systems would filter the bets as we moved down the systems.
Now, I was doing this a few weeks after the season started, so I hadn’t looked at the 2011/12 season at all yet. So, before releasing the backtested results and proofing the systems live (was done in the SBC forum), all I had to do was check the results last season were in line with previous seasons.
Errr…well, they weren’t. The first two months last season were losing months but more importantly, the type of bets thrown up during the backtesting didn’t exactly look like the same bets I was testing for during previous seasons. A quick look at odds portal and historical odds movements and I realised that the markets for these bets wasn’t fully liquidated on a Thursday evening (was like a first show if I’m honest with a high overrorund) and therefore, whilst I may have found an edge at kick-off prices, on a Thursday night, I wasn’t sure any edge was actually there.
So, I took the decision (first time I’ve ever done this which shows the lack of confidence I had in the systems) to not share any backtested results and I said I’d trial the bets in the SBC forum with no showing of the backtested results, so I suggested people see this as a paper trial. I started playing £5 stakes but only on Betfair every Saturday morning (and I’d take any odds available which were often much higher/lower than the Thursday quoted odds) and after seeing the systems do well for a little bit, I increased stakes to £10 just at the time the systems lost a fair bit! (I have a bad habit of increasing stakes on my own systems at the wrong bloody time!)
Anyway, as the season progressed last season, I could see that the results on the systems were nothing like backtesting and not only that, the number of bets thrown up was nothing like backtesting too. The Under system was finding more bets than during backtesting and clearly, it was diluting any profit the system could make. The Over system wasn’t a disaster but again, it was nothing like backtesting.
The other issue was the filtering of the bets as we moved down the systems. Some of the filtering was meaning that I was barely getting any bets thrown up and not only that, the bets that were being filtered out were sometimes the better bets!
Below is a table which shows the results from the systems proofed in the SBC forum last season. The systems are Under/Over 2.5 goal systems but I named them1, 2 and 3 to just confuse everyone! ;)
Overall, 920 bets and a profit of 2.8pts. Hardly a great endorsement of my system building skills for this particular market!
So, when I started looking at this on Sunday night, I had two options. I rip up the systems and start again (my preferred option) or I take a closer look at the systems to see if there is anything obvious standing out which is causing the poor performance last season.
My issue with starting again is that I would stumble across the identical issues I experienced with this system. I mentioned it when I built the Euro systems but there is a leap of faith in hoping an edge you find at kick-off prices is the same edge that is apparent earlier in the week. There is always risk involved in system building like I do but for the Under/Over system, it is quite big as I found last season. Odds ranges were a big part of the puzzle for the systems and ultimately, prices on a Thursday night weren’t anything like kick-off, so I was basically backing different teams than I wanted to back.
Now, there is a way around it I hear you cry. All I need to do is lift the prices on a Saturday at 2pm, plug them in my model and we’re off and running again surely? Well, maybe aye, maybe naw. I backtested last season with the kick-off prices and it was a rubbish season compared to previous seasons (about 8% ROI), so yes, it would maybe fix it a little but then again, do I fancy giving up my Saturday afternoon running footie models for an 8% return? Well, no.
So, I sort of decided quite quickly that my best chance of doing something was to look at the performance of the 6 systems from last season and decide if there is anything I can do to improve them.
Well, for starters, I need to bin the idea of filtering these systems. I don’t have enough bets on systems Under 1 and Over 1 at the moment and if I am going to filter these further, then systems 2 and 3 are already useless and will only get less useful as reduce the bets. Therefore, in the same way that UK system 9,10,23,24 and 25 went, the Under/Over 2.5 systems 2 and 3 will be consigned to system heaven. A system with less than 100 bets is no good to me (unless it’s a combined system producing massive ROIs) and therefore, going forward, we will only have a single Under and Over 2.5 goal system.
Right, let’s consider the Over system first as it’s nice and easy. Here’s the results for this system before it went live last season:
If I add the results from last season to the system, we now get this table:
Now, I gave this system a lot of stick last season and if I’m honest, I was maybe getting carried away with the fact the backtested returns looked as good as some of my other systems and I jumped to the conclusion that I may have uncovered a system on a new market that could give me an ROI of 15% (which I believe is very good in a market with such low average odds). However, it ended up with an ROI of 6.3% last season. As I’ve already said, all these bets were proofed in the SBC forum last season.
Is there anything wrong with a 6.3% return? Well, I guess not and therefore, I think it’s better that I leave this system as it is, let it roll on another season and see what happens next season. If we find last season was just a tougher season, then the 6.3% may increase but if we find it was a good season last season and it’s a rubbish system, the 6.3% return will decline.
Here’s the overall P&L and drawdowns graph for the Over system:
As for setting targets going forward, I’m not sure this is worth very much as I wouldn’t trust this system too much but I’ll give it a go!
I would expect we’ll see around 275 bets next season. Based on this season, I would hope we could get an ROI of around 6% again next season although I expect it will be in the range of -3% to +10% looking at the volatility this season.
So, that gives us a pts profit target of 16.5pts for next season.
Based on the drawdown graph, the max drawdown is 16pts. Hence, using the same idea as the other systems, I’d say a bank of 40pts should be adequate but given the average odds on the system are 2.08, it seems a big bank! Then again, I don’t really trust the system, so I’d rather have a bigger bank than a smaller bank.
The target ROC then becomes 41%. I’d think this is fair and if the system can repeat last season (which in all honesty, was rubbish), then I’ll be happy.
That leaves us with the Under system to look at then. Here’s the backtested results for this system before it went live:
A nice set of results but here’s what happens when last season is added in?
The first thing to point out is that although the system has done rubbish, similar to my other losing system last season (system 33), it is hardly a disaster. A loss of 2.4pts across 355 bets isn’t good but then again, I’ve seen plenty of systems perform worse than this one in a live environment, so I’m not going to beat myself up about having a poor system.
So, what do we do? Well, as I’ve already said, the issue is that the type of team I’m backing isn’t the same live as when I’m seeing them during backtesting and the first place to look at is the performance by odds. Lucky I started with this as it meant I didn’t need to go much further but it is usually the case you look at everything else first and then stumble across the answer last!
Anyway, if I strip out the bets at odds of lower than 2.00 last season and the previous seasons, it gives me the following results:
Now, the first point to make here is that I know that I am lowering the ROI overall by ignoring bets under 2.00. However, the intention here is to get something that looks reasonable when the system has went live and maybe it’s luck or maybe it’s something within the ratings but there are a helluva lot more bets at odds against last season than previous seasons and whereas good profits could be made historically backing odds on (better profits than backing at odds against), it hasn’t happened last season.
Secondly, last season, when I was backing these Under bets myself, I felt much, much more comfortable backing unders at odds against (e.g. going against the consensus) than backing at odds on as I felt like many of the odds on bets were just so obvious, I couldn’t see how an edge was apparent there. It’s a bit of a leap but I would be willing to bet that most of the Unders I was backing at last season were put in too short by the bookies on the Thursday evening and would have drifted considerably until kick-off. I’m not saying that backing odds on isn’t profitable but when the Under is put in at odds on a Thursday evening by the bookies, I’m not sure this is the same odds on selection I’d have seen during backtesting.
Thirdly, this is called backfitting something to the absolute max! I sort of believe there is a fundamental flaw on this system that kicked in before last season anyway (can see it in 2010/11 above for bets at 2.00+) and the bet number looked strange before last season, so I suspect there is maybe something underlying causing this.
Anyway, a lot of this is hypothesis as you can imagine and I’ve no real love for these Under/Over 2.5 goal systems myself (I won’t be following next season after my exploits last season), so I'm not going to worry too much.
So, looking at the table above then, this becomes the historical results for this Under system.
I would expect 200 bets next season and I think we should aim for an ROI of 5% based on the last two seasons. This gives a pts profit target of 10pts.
Here’s the results by month from last season so you can see what would have happened if you had been following both systems last season. Hardly a smooth journey and I think these systems probably aren’t for the faint hearted.
That’s another job done. Next job is to build the 4th UK rating algorithm which will be another DNB algorithm. I’ll be back in a few days with an update of how it’s going.