UBS using the Elo Ratings to predict the World Cup

Update 17 May 2010

Explanation for likelihood of making the second round. Check the end of the post.
______________________

Well, not only the Elo ratings...

Interesting analysis by Andreas Hoefert, UBS AG Chief Economist.

Source: UBS investor's guide - World Cup 2010 (starts on page 26).

In 2006, UBS correctly predicted Italy will win the 2006 World Cup. In addition, their model correctly predicted 50% of the semi-finalists, 75% of the final eight and 81% of the final 16.

Brazil - most likely champions: 22%, followed by Germany (18%) and Italy (13%).

South Africa - most likely team to reach the round of 16: 78%.

All group A teams have a likelihood of making the second round higher than 40%.

Korea Republic more likely to make the second round than Greece and Nigeria - same as my prediction of the World Cup.
__________________

David Hess raised the issue of the sum of odds (here and here) of making it to the next stage:

Either I don't understand what UBS is doing, or their predictions make no sense. Chance of reaching the Round of 16 for USA's group is listed as:

England 63%
USA 33%
Algeria/Slovenia not listed.

Given that the two teams of four make it, the total for the group should add up to 200%, no? Which means the odds for Algeria plus Slovenia should be (200 - 63 - 33 = 101). So Algeria and Slovenia advancing (101%) is more likely than England and the US (99%)? I don't think so.

Not to mention the odds for Group A add up to 221%.

I wrote to Andreas Hoefert and he was kind enough to explain:

The estimate was unconstrainted. Meaning I didn't put the constraint that the odds of each group should add to 200% in the first round. I did a contrainted estimate in 2006 with success and I did one again in 2008, which proved to be one if not the only cause for the miserable failure back then. Datamining suggested that uncontrained estimates delivered slightly better results without changing the rankings of team within a group. I.e. despite the fact that sometimes the odds of a group add to less than 200% and sometimes to more, the relative ranking within the group, i.e. England is more likely to advance than the US, which is in turn more likely to advance than Algeria, which is in turn more likely to advance than Slovenia.

Hope this helps.

Kind regards

Andreas

---

About me:

Christian, husband, father x 3, programmer, Romanian. Started the blog in March 2007. Quit in April 2018. You can find me on LinkedIn.