Re: World Cup Simulation Results Ok, let's suppose somebody *had* put money on these four teams. According to Shackleton's calculations (post #4 of this thread), the pre-Cup odds suggested their combined probability of winning the Cup was 16.9%. As of now, that figure has risen to ~30%, so a cautious bettor could presumably guarantee himself a profit by hedging. Of course, if Brazil had won today, things wouldn't be so happy.
Re: World Cup Simulation Results Was thinking of this post today and really really wishing I'd done it, with Germany added to the mix (Germany added a very very small edge over the betting odds, according to Voros, and would have given you a ~56% chance of winning overall).
Calling two out of the four for the final four is not bad. How about run it again with now the known final four and see how the percentage changes?
Re: World Cup Simulation Results Hehe - your post actually did inspire me to bet on France. I doubt it will still happen, but a very small bet on my part may pay off pretty well. Thanks! (And thanks to voros.)
A brief initial post, I'll go in to much more detail later. Using the numbers I posted on the web before, of the 49 non-drawn world cup games, the system picked an incorrect winner in 7. 1 of those 7 was drawn after extra time (Italy v Germany). The other 6 Poland v Ecuador Ghana v Czech Republic Ghana v U.S.A. Italy v Czech Republic Netherlands v Portugal France v Brazil As numerista points out, the end result is a .773 win% for the system picked favorites when the system only predicted a .655 win% for the favorites. This has been a fairly common problem in the system and has been difficult to correct. Initially, if I had to point a finger, I would point it squarely at poisson as a possible culprit. But the initial tests I've done suggest that isn't it, and previous tests on data have shown a gap much, much smaller than that from this WC. So I think to some extent that result is anomalous. My guess is that last World Cup's results would be less so. I think maybe the use of uncompetitive matches might also be partly to blame, but as I explained using them has tended to be a far better solution than not. Maybe the 'European benefit' discussed before might be to blame and if I had boosted the chances of European teams higher, the expected win% would be higher. I could rejigger the numbers post rating, but that seems kludgy to me. For ELO favorites, predicted was .694 and actual was .742, so they had the same problem (though certainly less of one). Also they lost the same 7 games my system lost, plus they lost Costa Rica v Ecuador and Ivory Coast v Serbia. The good news is that this shouldn't effect the team _rankings_ as much as the _ratings_. Team A is still better than Team B, but maybe the gap ought to be bigger than the ratings show. Also, using an NFL pool type system (assigning a point value to each match from 1 to 64 with the largest points going to the biggest mismatches), the system outperformed FIFA by a country mile and ELO by an amount I'm not sure what to make of (1718.5 to 1672). Total possible points were 2080.