Collaboration in Forecasting

Abstract

Collaboration and information sharing is becoming an increasingly important tool in improving forecasting socio-political events. How can we harness the power of crowds for forecasting without its associated limitations? In this paper, we explore how different types and amounts of information sharing affect forecast quality. The results of the experiment show that, overall, different types of information sharing improve performance over no sharing. In addition, information sharing does not appear to harm aggregation weighting methods that depend on meta-predictions about group performance .We discuss implications for these results for use in improving forecasting.