Forecasters increasingly are employing sophisticated tools to analyze vast volumes of data to make predictions. To test the latest forecasting technology, New Scientist recently recruited teams of researchers to predict the magazine's future sales.
New York University researchers Max Sklar and Matthew Rathbone started by identifying and extrapolating long-term trends in the magazine's sales. They developed a method for adjusting the predictions based on the seasonal variation in the sales, and the team's prediction was within 1,000 copies of the actual figure.
George Mason University's Robin Hanson developed a prediction market in the 1990s that relies on collating human judgment, and he set up a prediction market with the New Scientist staff. The staff's collective decisions would drive the share price of an issue up or down, and the closing price each week was used to predict how well that issue would sell. Hanson's prediction market technique got results similar to the New York University team. Finally, the New Scientist used Amazon's Mechanical Turk method to commission workers to predict the magazine sales. The "turkers" were no more accurate than the other methods.
From New Scientist
View Full Article