Masters of Unpredictability

When I was watching BBC’s Match of the Day 2 on Sunday evening, the programme host mentioned how unpredictable Aston Villa’s league form has been lately. He pointed out that their last 7 league game results (before last Sunday) were: loss-draw-win-loss-draw- win-draw. It seems indeed a little random in a sense that the team hasn’t achieved back to back wins or draws or losses. After last Sunday’s game, a 2:1 loss to Newcastle, adding to the sequence, the hypothesis still holds.

But are they really being unpredictable?

Let’s say Villa were up against Man Utd, Sunderland and Wigan in the first three games. Then a result of loss-draw-win wouldn’t be much of a surprise anymore. So when is a team unpredictable? I’ve decided that it’s when a team is expected to lose a game but manages to win it and vice versa. This means we need to account for the opposition’s team strength. A team of course could not be “unpredictable” all the time (against expectation, derived from our Dectech model), otherwise it means our model is rubbish (which can’t be!) and the bookies are going to make a huge loss.

I want to look at which teams have been unpredictable this season. My idea of measuring them is to look at, on the match-by-match level, the squared difference of probability of most likely match outcome (either win or loss or draw, I’m not looking at the number of goals) and the probability of actual outcome. A seasonal average of such measure by team can reveal how unpredictable a team has been. Let’s take an example, when Man Utd were playing against Blackburn during Christmas, we predicted a 78% chance of a Utd win and 9% chance of Blackburn win. So if Man Utd had won the game, the unpredictability of that game would be: (78%-78%)^2=0. On the other hand, if Man Utd lost to Blackburn (which they did), the unpredictability measure was hence (78%-9%)^2=0.48. The good thing about this measure (I believe) is that it also reflects how shocking the result is – if a game’s Win/Draw/Loss probability is 40%/30%/30%, then even though a loss is not expected, it’s still not that unexpected, relative to the most likely outcome ((40%-30%)^2=0.01).

Time to show some results: below are displayed the 10 most unpredictable results so far:

Applying the unpredictablilty measure to all 240 matches that have been played in the Premier League this season, we have the most unpredictable teams:

Chelsea, the team who have shared 4 out of the top 10 most unpredictable results, not surprisingly sit at the top. Perhaps you could call it disappointing rather than unpredictable as too often this season they have failed to win a game which they should have. Both Liverpool and Arsenal probably share the same rationale for being at the top. Wolves are probably the most genuine unpredictable team here: they have geo points when they weren’t expected to (draw to Tottenham and Arsenal away) and lost games when they should have won, a mixture of both joy and frustration.

Finally, I’ve taken a look at the relationship between team strengths and their unpredictability and below is what I found:

From this season’s data we cannot really see any clear trend. However, by scrutinizing some individual teams we can get something interesting (if you like betting): for example we reckon Chelsea are still a slightly stronger team than Man Utd, but Chelsea have been behaving rather unpredictably this season so far. This tells us that it might be worth betting against Chelsea rather than Man Utd given the same profit edge. As for Man City, a team predictably doing well, is their investment finally going to pay off by winning the league this season?

One thought on “Masters of Unpredictability

  1. If you’ve looked at previous years,are you seeing teams that are unpredictable in the first half of a season being similarly unpredictable in the second half of the season.You might just be seeing small sample size randomness.

Leave a Reply

Your email address will not be published.