Home From Home
Home From Home
If you ready my previous article, you would have been aware that the Bundesliga undertook a rare round of midweek fixtures over the last two nights and where there is a gameweek, there is a chance for data driven betting excellence.
This meant it was time for ‘BundesBot’ to once again try his luck and after returning four out of nine data driven football predictions last weekend, the quest was to return to the more impressive seven in week one.
With a difference in performance, it is hard to ascertain, which was the fluke and which was the more standard measure and therefore, something that I’ll get a greater idea of once we review the midweek results.
To do so, here’s a look at what ‘BundesBot’ served up on Tuesday afternoon:
After Bayern Munich got the better of Borussia Dortmund in Der Klassiker, it saw the midweek action off to a perfect start. However, that is about as good as it got, because after that it was an absoulte disaster.
The only other correct result after that, was Augsburg’s draw with Paderborn and this meant that out of nine data driven football predictions, only two came back as correct. You could sum this up by relegation form and not a lot more.
With that said, one must not forget that the odds for this nine-fold to come good were absolutely ludicrous, 38,000-1 (and more), which meant one of two things:
- I was going to be very rich
- I was going to be very wrong
Unfortunately, it turned out to be later.
Results such as Bayer Leverkusen getting pumped 4-1 at home to Wolfsburg at home certainly did not help and although ‘BundesBot’ went a bit more leftfield with a predicted draw at the Bayer Arena, it (or anyone else for that matter) did not see this one coming (more on this in a minute)
This means that the performance of ‘BundesBot’ has decreased in the three rounds of fixtures, with the return now as follows:
Week 0 (Control) – 7/9 (77.7%)
Week 1 – 4/9 (44.4%)
Week 2 – 2/9 (22.2%)
Total – 13/27 (48.1%)
The hit rate is now less than half, which suggests that gut instinct could probably do as good or a better job a present. Although without being too hard on myself, there are perhaps some mititgating cirumstances.
WHAT HOME ADVANTAGE
After those same 27 fixtures that have been played, just 5 have resulted in a home win – this equates to 18.5% (less than one in five) and it is element of no home advantage, which is scuppering my hopes of winning it big.
If the notion of home advantage has now been removed, it means all games for all games clubs are winnable. No longer, is the viewpoint that a win will more likely come every other week, everything up for grabs.
Which is great if you are a manager of a Bundesliga outfit, because the belief within the camp in each game will be as high as it can be possibly be and there’s no longer a sense, “we might be able to nick something here”
With that said, its an absolute nightmare from a betting perspective, because all the usual and previous intiuation you may have had is now simply a thing of the past – something which makes predicitve modelling a huge crap shoot.
At the same time, this is complete random either. For example, Bayern Munich have played two of their three restart matches away from the Allianz Arena and the likelihood is that they would win where they play.
Then again, you only have to look at midweek to see the slips by Bayer Leverkusen and RB Leipzig, to see that the wisdom of crowds (or lack thereof) is certainly having a bearing on predicted outcomes.
WHAT TO DO NEXT
Well the obvious answer, is to do far better than two correct out of nine and it does seem that four or five out of nine correct is the benchmark, with wild swings in performance on either side. Therefore, everything is still rather erratic.
This will mean another round of testing this weekend, but I think there will be a first wave of changes to be implemented. As to what they are at the moment, I’m not too sure, but I can’t see adding additional weight to teams at home being the solution.
The next step may be some A/B testing and testing the current framework against fundamental change, otherwise you end up changing changes and losing track of actually works and what does not.
However, we can come to the simple conclusion that last week was a disaster, but a hint of that is based on the Bundesliga being turned on its head in terms of the results that everyone had expected.
IT’S NOT ALL BAD
Still there is some light to shine and that is not the double chance bet’s that I placed over the midweek section, a 9-fold combo that looks as follows (incorrects in italics)
Bayern and Draw
Leverkusen and Draw
Freiburg and Draw
Monchengladbach and Draw
Hertha Berlin and Draw
FC Koln and Draw
Schalke and Draw
Mainz and Draw
Augsburg and Draw
A half decent (or should that be two-thirds?) offering as the ‘BundesBot double-chance bet, returned six data driven football predictions out of nine. Which is the same return, as the weekend before.
Out of 18 double chance bets, 12 of them have been correct – a 66.6% hit rate and it makes me think, this might be the better route in which to tread upon. However, this also brings its own issue when it comes to making changes.
Currently the model is performing well in terms of double chance but in terms of actual result, but the latter is connected to the former and therefore, any changes made to ‘BundesBot’ might knock out the success of the insurance policy.
This is something that I need to keep in mind, or the other angle is that I simply make no changes at all and just say ‘this is what I’ve built’ – can it end up being a success, that though is no real fun and a bit lazy.
In summary: It’s been a bad midweek in terms of overall predictions, but double chance looks promising. Now I need to work out what needs to be done and in quick time, as we go again on Friday night.
Happy punting and thanks for reading. Dan
(THESE ARE NOT TIPS PER SE, THESE ARE JUST DATA DRIVEN FOOTBALL PREDICTIONS IN A PURE TEST AND LEARN MODELLING CAPACITY)
If this has grabbed your interest and you would like to discuss/feedback then please feel free to drop me a message at firstname.lastname@example.org. While I am always looking for new football/data projects to work on and if you feel that my skills would be of use, I can be contacted at the same address.