Who Won the Olympics Forecasting Competition?

At least three sets of economic researchers projected the medal count, by country, before the Games began. Two of them shied away from the word “prediction,” even as it was attached to their reports. Perhaps not surprisingly, the third forecast came closest to the actual final medal count, just as it did in 2004.

The three forecasts had a lot in common. All three – from Colorado College, Dartmouth's business school and PricewaterhouseCoopers – used models incorporating GDP, population and a host-nation effect. And all three underestimated U.S. and China medals – by an average of 11 medals and 14 medals, respectively – while overestimating Russia's haul, by an average of 17 medals. Among countries they all forecast, though, Dartmouth had the closest fit.

The forecasters at PricewaterhouseCoopers and Colorado College say they didn't set out to predict medal counts, though media accounts of their work don't usually make this distinction. Instead, their goal was to set benchmarks that represented a par performance; countries exceeding those benchmarks could be seen as having successful Olympic Games. "If we were doing a forecast, we would try to do each sport," as Sports Illustrated does, John Hawksworth, head of macroeconomics for UK PricewaterhouseCoopers LLP, said. "We were trying to approach it much more from a top-down perspective: what factors have influenced past performance and therefore what would be a reasonable benchmark." Nonetheless, his benchmark incorporates the medal share from prior Summer Games, which means that a country that surpassed expectations last time has a tougher go of it this time.

Daniel Johnson, an economist at Colorado College, doesn't include prior medal shares, noting that his model is "as purely economic and socioeconomic as possible." In fact, he hasn't tweaked it since it was first introduced in 2000. "We didn't think of doing predictions," Mr. Johnson told me, though his team ultimately decided to release its analyses ahead of the Games, as "an interesting check of whether our model holds up over time, or is just good at explaining the past."

Read more…