Evidence on good forecasting practices from the good judgment project: An accompanying blog post

Abstract

The Good Judgment Project (GJP) conducted a forecasting tournament from 2011 to 2015, where thousands of online volunteers predicted geopolitical events. GJP identified the top 2% of predictors as “superforecasters” and studied their forecasting practices, comparing their performance to various control groups, including expert political judgment, prediction markets, and simple algorithms. GJP found that superforecasters outperformed all other groups, demonstrating that forecasting accuracy is not only a matter of luck but also of specific skills and habits. Superforecasters demonstrate an open-minded cognitive style, are comfortable with numbers, and engage in rigorous belief updating. GJP also conducted a training experiment to test whether forecasting ability could be improved through training, finding that a one-hour training module consistently improved accuracy. The article analyzes the findings of this research and concludes that the skills and habits employed by superforecasters are likely applicable beyond the domain of geopolitical forecasting, suggesting that these findings can be applied to other fields, including the prediction of future technological developments. – AI-generated abstract