A Downward Spiral

A Downward Spiral

If you read my previous article, you will be aware that it is once again time to attempt some data driven betting excellence and with another Bundesliga weekend now in the history books, it is time to see how ‘BundesBot’ has fared.

It’s been something of a rollercoaster in terms of performance for this particular predictive model and with the number of correct data driven football predictions ranging from as little as 2 or as many as 7, its hard to hang your hat on overwhelming success or abject failure.

Obviously the end game is to get all nine per week correct (or 10 when we pivot to the Premier League this time next week – even more juicy content!) but at the same time, it is all about guaging the level of success and then trying to work out if a predictive model can be considered succesful.

It’s last showing offered up seven correct predictions from nine, meaning I was just two away from earning a 585/1 accumulator. Unfortunately, this just fell short and it also means, it was time to try our luck once again.

HOW DID WE DO?

Take a bow if you are:

Bayern Munich
Wolfsburg
Mainz
Union Berlin and Schalke

Although to offer some protestation, there were some undoubted slip ups – Borussia Monchengladbach and RB Leipzig I am looking squarely in your direction, after you both failed to win this weekend.

Still, it is no good me bemoaning the shocks, because you can only work with you’ve got. Although if there is a crumb of comfort, it’s that ‘BundesBot’ managed to pick up some of the more unexpected result (Mainz Win, Union/Schalke Draw)

So it’s almost as if it did the hard work and was let down by some shock results. With that said, it did only return four correct predictions out of nine and this means we’ve taken a bit of a hammering in terms of the success rate.

If we breakdown the week by week success rate:

Week 0 (Control Week): 7/9
Week 1: 4/9
Week 2: 2/9
Week 3: 7/9
Week 4: 4/9

Which means as a whole, 24 out of 45 correct results have been picked (53.3%) – more than half and in the face of some freak results along the way (especially in terms of a lack over home advantage since the restart)

Of course, I always going to be on a hiding to nothing with this set of data driven football predictions and if only for the simple reason that the odds of this nine-fold coming in was a rather hefty 3,496/1 – something that was almost impossible from the start.

The big jump in odds, was due to ‘BundesBot’ backing Hertha Berlin to win at odds of 7/1 and this multiplier was always going to scale things up considerably (while not to mention being the wrong outcome at full time)

This meant with the two shocks mentioned above and one rank outsider being selected as a prediction, it put a huge squeeze on the overall outcome and although the overall hit rate is more than half, there is a sense that more can still be done.

Avid readers (of which there are some) will be aware that I’ve not made any changes to the model in the first five gameweeks of its operation and there are two schools of thought to finish the season:

1) Let it ride out
2) Change and see what happens

If I was to run this over 34 weeks of the Bundesliga season, it would be incredibly stubborn of me to build a predictive model and leave it unchanged, especially as I would be picking up new nuggets of information along the way.

However, with this being a short-term blitz of bets and there has also been two very impressive returns, it does perhaps suggest that we should let this model build run for the final four gameweeks of the season.

I think the target will be to beat 50% success and at least use that as a benchmark for next season – although of course, there almost needs to be a 34 week model run tested against another one and therefore, the real success might not come a couple of seasons down the line.

With that said, there’s patience and time in abundance, so no harm at all in testing with new variations and next season will be the campaign in which all production and ideas are ramped up considerably.

TAKE A DOUBLE CHANCE

Once again the double chance market seemed relatively impressive in terms of data driven football predictions and from the four gameweeks that have been served up, the returns have been 6,6,8,6 respectively.

This means that from 36 matches, 26 of them have been correct in terms of the double chance – giving me a 72.2% success rate. Although, the caveat is that the odds are far lower in this and even if you won a nine-fold, it hardly seems the overall effort.

However, it does also suggest that ‘BundesBot’ is not completely out of whack and is getting hampered more by draws than the actual winner of the game, something that suggests the draw mechanism needs adjusting.

If I was working on this all season, it is a change I would make but I’m going to let it ride and see if data driven betting excellence is not simply around the corner, a corner that will appear on the horizon later in the week.

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 dan@realfootballman.com. 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.

Follow me on Facebook at Dan The Stat Man
You can also check out my Premier League Podcast on Soundcloud