This paper combines survey forecasts with a heterogeneous agent model to examine the dispersion of expectations of participants in the foreign exchange market. We find distinct variations in the level of dispersion and document that dispersion arises because of the combined effect of market participants holding private information and attaching different weights to fundamental, technical, and carry trade analyses. We estimate a heterogeneous agent model on the survey forecasts and show that the weight attached to the three forecast rules is adjusted over time in response to the relative importance of the rules in the actual foreign exchange market. The weights are related to market circumstances; the switching model is finally shown to outperform the random walk model in an out-of-sample forecast exercise. © 2012 Elsevier B.V..