Abstract/Summary

We show that typical tests of whether forecasters herd will falsely indicate herding behaviour for a variety of types of behaviour and forecasting environments that give rise to disagreement amongst forecasters. We establish that forecasters will appear to herd if di¤erences between them reject noise as opposed to private information, or if they arise from informational rigidities. Noise can have a behavioural interpretation, and if so will depend on the behavioural model under consideration. An application of the herding tests to US quarterly survey forecasts of inflation and output growth data 1981-2013 does not support herding behaviour.