Friday, July 1, 2016
Where were the superforecasters?
Predicting methods are among the losers in the Brexit referendum and in the Spanish election. The night before the referendum I confidently sent an email to some friends reassuring them that betting and prediction markets were safely forecasting a remain victory, almost for sure. My apologies. The Economist immediately wrote an article after the referendum claiming that polls were better than prediction markets, because the latter were victim of all the behavioural biases that affect financial markets. However, four days later, polls also miserably failed to predict the Spanish election. Poll results published the night before the election (as well as those published during the campaign) predicted that the Socialist Party would be clearly overcome in votes by the populist left, and that all the leftist parties together would have more seats in Parliament than the sum of the right and the center-right. Even in the election day, exit polls made a similar predicition, only to be ridiculed by the real results a couple of hours later. With the benefit of hindsight, pundits have concluded that the Brexit result had a significant impact on a decisive part of the electorate: conservative voters rallied around the Popular Party and the populist left lost some cautious voters. If this is true, this would not be an explanation for the polls from Friday to Sunday. The impact of the Brexit result on the Spanish election is not a random issue. The Brexit result was a clear possibility, and once it took place, its impact on voters was not something that could not be explained. The political chaos that followed the referendum in the UK was a warning against those parties that promote plebiscites and that do not have a clear commitment to the European Union. The referendum provided information (a public good) that was relevant in Spain. The consequence was that the populst left would lose a significant number of votes (other were lost during the campaign for other reasons). Both polls and predicition markets got it wrong. Next time we should mobilize Tetlock's superforecasters.