If you read my post just before the Euro 2016 started, you might be wondering how my model actually did. Well most likely you have not been wondering at all about that but I have decided to share with you how i did anyway. I did not do too bad. But before I go into the results, in three built points, let me remind you what my model actually did:
- First, using the historical relationship between goals scored at a major tournament and the Fifa rankings I came up with a theoretical most likely goal difference for each game in the group stages of the Euro 2016;
- Second, I used this theoretical goal difference to calculate the probability of the three possible outcomes of each game (see for example my prediction for France Romania here);
- Third, I calculated the expected value (likelyhood of outcome, times the payout less the cost of the bet) from every single possible bet based on the odds of offer at Betfred
Using this method I had calculated, even if my probabilities had been perfectly accurate, that I could stand to win north of £100 and south of zero with the most likely outcome being around £20. Obviously, given that I was working with very imperfect data the true expected value of my bets was likely to be well below that of £20.
But how much did I actually win/loose
Getting to the point, it turns out my model did not do to bad in the end. In fact, my model lead me to beat the market by £1.68! Now if we discount my own opportunity cost this is a sizable 3.4% return on my investment. But as the figure below shows, at times I was not confident that I could boast of a positive return on my investment.
Very much like the games of the tournament itself where more than 60% of all goals came in the second half of matches, my model only picked up during the second half of the tournament. The fact of the matter is that I was pretty sure I was done and had no way of coming back when Italy beat Sweden and I had put a pound on a draw.
Figure: Saved in the second half
But, thanks to some high grossing results such as Czech and Austria drawing respectively against Croatia and Portugal my model picked up and so did my cumulative net position (green line on chart above) improved. Then I was save after the epic 3-3 draw between Portugal and Hungary coupled with Belgium beating Sweden. At that point I knew I would come on top.
The lesson I learned – or what have become more determined in my view
Firstly and obviously, without any additional information to the market and no expert knowledge of the game, it is incredibly hard to beat the betting market. After all the collective wisdom of the betting markets for football is many times greater than mine. But what I think helped me make this tiny bit of profit comes down to two points:
- First and most importantly, thinking in probabilistic ranges rather than absolutes. This, in my view, is the crucial ingredient in any form of prediction making. It is absolutely crucial to understand that 1% chance is still a chance. in fact if you were to make a call on an outcome with a true probability of 1% 69 times there is a 50% chance that in one of those instances you were correct.
- Second, odds are not enough, in order to make a successful bet you also need to take into account others overconfidence in a certain outcome. This does indeed relate to the firs point – if you think probabilistically and others think binary, then if you bet on sufficiently large number of games then you have a greater chance of coming out on top. This is where the concept of an ‘expected value’ proves to be an effective decision making tool. But, I do not bore you with the details, but for those interested, you can see by looking at the original bets I made (in the annex of this blog) that only 30% of my bets were for the favorite.
But now the football is long gone, I have just under £52 on my Betfair account and the Marathon at Rio is taking place on Sunday. Stay tuned.