How do firms adjust their expectations as new information arrives? The response of forecast errors to news provides useful insights: if firms process news immediately and correctly, news will not predict forecast errors. Yet we find that micro news, that is, news regarding firm-specific developments, give rise to negative forecast errors. Instead, macro news induce positive forecast errors. This implies that expectations overreact to micro news, but underreact to macro news. We establish these results for a large survey of German firms and put forward a general equilibrium model with dispersed information and overconfidence to rationalize the evidence. In the model, overconfidence raises the output dispersion across firms but dampens aggregate fluctuations.