We study the expectations of individual forecasters in the foreign exchange market. We find that the survey risk premium is less countercyclical than the rational risk premium primarily because it is not related to the forward premium. We also find that forecasters learn from their own forecast errors (rather than from consensus forecast errors) and that they overreact when forming expectations (as indicated by their forecast revisions). Finally while forecasters have worse forecasting performance relative to a simple benchmark the forecasters who emphasize the real exchange rate and do not overreact have better out-of-sample forecasting performance. Overall our results highlight the information contained in individual (rather than consensus) exchange rate forecasts.