The Goldman Sachs model was already discussed here. I wonder whether there are other statistical forecasts publicly available, or even better, models including raw data. Specifically, I thought that the Goldman Sachs model didn't work so well for the 2010 cup (see the original report on page 7). I also heard that actual or estimated transfer values of players have high predictive power, but they are not included in the Goldman Sachs model.
I like this approach: http://www.r-bloggers.com/home-victory-for-brazil-in-the-2014-fifa-world-cup/ (more formal description in the working paper here: http://econpapers.repec.org/paper/innwpaper/2014-17.htm)
The idea is to use a Bradley-Terry model where the team ability parameters are found using simulations of whole tournaments in order to match bookmaker probabilties of each team winning the tournament.
I suspect even better results could be achieved by using the implied probabilities of the Betfair market.
I haven't tried it personally but there is this application, by a company name Bayesia, that is relevant to your question: http://worldcup.bayesialab.com/.