Some Miscellaneous Thoughts on European Football Picks So Far This Season

This post is partially prompted by the frustrations some are having with the European football picks so far this season and also partially by last Saturday's stream, in which Robert hinted at paring back the football picks for now and focusing on a smaller number of targeted long shots. I believe it's very important to put association football betting into perspective as it's a totally different ball game (literally) to American sports and so we need somewhat different tools to crack the nut so to speak. Fair warning, this post will meander for quite a bit, but I am going somewhere with it.

As I've alluded to in previous posts, there are two primary differences between European and American sports. The first is that European sports tend to be more strategic, orientated around the ebb and flow of real time play, whereas American sports are much more tactical and built around key set pieces. The second difference is that European sports very often end in draws, even without the offering of extra time. This latter difference is the crucial challenge from a betting point of view as the underlying draw probability (and hence fair value draw odds) have enormous implications for the profitability of a strategy.

The standard deviation of a binary fixed probability outcome is equal to the square root of the product of success and failure probabilities. The standard deviation of the same over a number of iterations is equal to the standard deviation above multiplied by the square root on the number iterations. Although the standard deviation of your dataset rises as the number of iterations rises, the standard deviation as a fraction of total iterations actually goes down. This is the foundational principle behind the high volume betting strategy that makes up the official picks.

How do we compare the variances of two different binary outcome events, one with probability close to 50-50 and another closer to 30-70? The standard deviation of the former over one iteration is equal to 0.5, whereas the standard deviation of the latter over one iteration is equal to about 0.46, which is actually a bit less than the straight 50-50 coinflip but due to one outcome being less likely than the other, the iteration to iteration variance is bigger. In the long run, long shot bets have less variance but over smaller sample sizes, their results are much more erratic.

I was slightly surprised when I heard Robert say he may scale back on the football picks somewhat. Over the weekend we had a cluster of long shots hit that were very much consistent with the feast or famine pattern that he advised us to be prepared for. Nonetheless, I have been tinkering with a couple of unconventional angles that might help us reduce the famines and increase the feasts.

I'd like to note that official baseball picks have been extremely successful this year so far; yet if you read the comment sections on Polymarket (or if you dare venture into the caverns of Reddit), you will see them full of people tearing their hair out over baseball betting. The common complaint is the ridiculous amount of variance, teams throwing away leads at the death and the high amount of play attempts just to register a score. Furthermore, I have also learned that the official baseball picks used a different system several years ago and that the current system made some significant conceptual changes which led to its present profitability. On top of that, I'm aware that conceptual changes have been made to the American football betting systems used in the official picks too. I suspect a similar conceptual change may hold similar power for the association football betting system.

Over the last few years, football matches have come to be dominated by longer and longer periods of injury time. What used to be three or four minutes added on to the ninety is now regularly more than ten. If you look at live odds going into injury time and the two teams are close in score, there'll often be implied chances of at least 25% for a game changing goal to be scored. Even in Manchester United's heyday with the so-called Fergie Time this would have been considered excessive. What if we instead of modelling attack versus defence during stoppage time we treat defences as if they were the same and that we give the edge to whichever team has the better attackers? Last season we had countless bets that were on track to win crash out for a miss due to last minute goals. This might be an opportunity to finally turn this in our favour.

I have noticed that the official picks have not been recommending quite so many draws lately. In order to assess which position to take, I have been considering estimated win, lose and draw probabilities and calculating expectation values from them. For this reason, the good to ranges quoted in the picks are of tremendous importance. Whenever a pick gives a slight edge to an underdog relative to their quoted odds, I have found better luck buying no on the favourites (I can normally take advantage of better liquidity). When the underdog's good to odds are considerably better than quoted, this implies a very significant edge; however for more evenly matched teams, my suspicion is that the draw is more often than not the sensible option. The crucial breakthrough I had when backtesting last year's picks was turning all the pick'ems into draws. I couldn't apply this to every single pick'em in the top leagues, but for whatever reason, it worked beautifully for the official picks.

I'll let you in on a little secret of mine: when two big market teams play each other and are equally matched, you can often get better value than a straight draw by taking out no positions on both to form a synthetic draw. Due to order book liquidity, you may find it difficult to find someone who will sell you the draw but you can easily find someone who will buy the win for either side from you. Because of this, I managed to get Man Utd vs Arsenal at considerably better odds via the synthetic draw as opposed to betting on them to draw from a bookmaker. Unfortunately United couldn't hit a barn door that day but I think on another day, it would have worked out really well.

I think the value in football betting will be selling big market teams against mid market teams in a manner similar to baseball with potentially a lean towards mid market teams with potent attacks or selling mid to big market teams against each other to form synthetic draws, especially when said teams do not have strong offences. It may well take combining the statistical analysis of baseball with the sharp in advance line shopping of NFL but I do suspect more and more that football is entering a new phase of prediction modelling. With prediction markets at our fingertips this will be an exciting year yet.