World Cup forecast should be taken with a grain of salt
The Olympics are the world's biggest sporting event, but they are a long way from being the most exciting. Apart from a handful of devoted enthusiasts, no one really cares about the results of the 10-metre air-pistol shooting or who wins team gold in synchronised swimming.
No, when it comes to whipping up global sporting frenzy, the football World Cup is in a league of its own. In just 42 days, Germany will take to the field in Munich against Costa Rica, kicking off a month of football mania.
Trading volumes on international financial markets will fall, internet bookmakers will take billions of US dollars in bets and, across Asia, millions will stay up late into the night to catch their favourite teams on television.
The identity of the eventual champions who will lift the trophy in Berlin on July 9 will be the subject of intense speculation.
Actually, it already is. In an attempt to get in on the act, Andreas Hofert and his team of investment strategists at UBS Wealth Management in New York have applied the sort of numerical modelling they normally use to determine stock and bond market allocations to this year's World Cup. Their forecasts could raise a few eyebrows.
The most probable winner will be Italy, beating out the bookies' favourite Brazil in the final, Mr Hofert believes. If that sounds farfetched, altogether more realistic is the lamentable performance Mr Hofert expects from second-favourite England. He predicts Sven-Goran Eriksson's team will crash out ignominiously to the Netherlands in the quarter finals.
If all this sounds like a flippant exercise in marketing fluff, it is. But it is also a telling illustration of the strengths and weaknesses of financial models.
Like the trading systems devised by hedge fund managers to play financial markets, Mr Hofert's model relies heavily on past data. In this case, the UBS analysts have combined several historical data sets, including the number of times teams have reached the semifinals, their Elo ratings, which gauge teams' recent performances, and the number of outstanding players in each team, according to Pele's 2004 rankings.
The model is then weighted for factors which have a statistical significance in past results, such as home country advantage.
The UBS analysts claim impressive accuracy in historical back-tests, saying their model correctly predicts 70 per cent of quarterfinal winners in the past five World Cups and 89 per cent of semifinal winners.
Inevitably, there is a certain amount of 'curve-fitting' going on here. It should not be surprising if the model correctly forecasts the results of past competitions as that is exactly the data that went into the model in the first place.
This is a problem with trading systems, too. They always perform strongly in back-tests but often less well in real life.
In the end, Mr Hofert's model is only a loose probabilistic guide. His forecast that Italy has a 54 per cent chance of beating Brazil in the final should be taken with a pinch of salt. After all, if the model is run cumulatively, rather than started afresh for each round of the competition, Italy is still the favourite but with only a 9 per cent probability of emerging as the victor. That means there is a 91 per cent chance the winner will be someone else.
Still, it is hard to imagine anyone performing a similar exercise to forecast the gold medallist in the modern pentathlon.
