Prediction markets are virtual stock markets, which use the information contained in market values to make forecasts. As a group of Nobel laureates has emphasized (Arrow et al. 2008), prediction markets belong to the most potent forecasting methods. Prediction markets have proven themselves in the prediction of elections in the USA (Berg, Nelson, and Rietz 2003, 2008) and been widely used in Europe to predict election results where they have outperformed other forecasting models based on polls, expert panels, and economic indicators (Graefe 2017). Even more, recently prediction markets have successfully been applied to forecast highly specific events such as replication studies in social and behavioral science (Gordon et al. 2021) and migration flows (see below).

We do not only apply, but also further develop prediction markets. For instance, in the context of the project “A Prediction Market with Integrated Algorithms (PREMIA)” (funded by the Digital Society Initiative of the University of Zurich) Oliver and colleagues develop trading algorithms that might improve the accuracy of prediction market forecasts. Also, together with Sandra Morgenstern (University of Mannheim), he has conducted a prediction market and evaluated its accuracy for the tough case of migration movements. Before, Strijbis has studied the accuracy of the prediction market for direct democratic votes.

Usually, we publish raw data from the market as probability forecasts. However, in some instances such as vote or seat shares of parties we first interpolate probability estimates for different ranges (e.g. ranges of vote shares) in order to arrive at a point estimate and a confidence interval. Also, when combining these forecasts for different parties in one election we correct in case that the forecasted sum of vote or seat shares deviates from 100%.