[Dash2in1]

I'm wondering if it's possible to calculate an approximate solution to the following question:

"Suppose" there's a fantasy league for the upcoming Euro 2012, with a number of players, each predicting a scoreline for the group matches. Suppose further that you have a distribution of scoreline for every match and that previous results are uncorrelated with next results (which is untrue, but it's much, much tougher otherwise).

Points are given as follows, exclusively (!):
- 5 points for a correct score
- 3 points for a correct goal difference (i.e. predicting 2-0 and the actual score is 3-1, you get 3, not to be confused with the scoreline being 1-3, you get 0 for that)
- 2 points for picking the correct winner or draw.

Now you could easily calculate the expected point value per match given the score distribution. However, doing that will not give you the optimal solution. So, is it possible if you have a distribution of scorelines and the predictions of all other players and how to do it?

[Dash2in1]

Something that comes to mind is the following:
- always take the highest point value play (which in this case probably most of the time (or always) is 1-0 or 0-1, given that you also cover scores like 2-1 and 3-2 for the favorite for 3 points and 2-0,3-1, etc and draws are less frequent outcomes)
- if for instance sufficiently enough people take the 1-0, 2-1, etc for the favorite, at some point it becomes favorable to pick 0-0 or 1-1.

Question then becomes: when is it enough to switch?