In this paper, we characterize a forecasting model where forecast ers cannot perfectly distinguish between the two persistent components (trends and cycles) in a dynamic setting. In this model, forecasters jointly update their beliefs about the two components: noisy information about one component is used to update beliefs about the other component. We present diagnostic empirical facts on forecasting behaviors and show that these facts are consistent with our model’s predictions while contradicting those of existing models in the expectation formation literature. To vali date our model, we exploit the Federal Reserve’s 2012 adoption of explicit inflation targeting as a policy shock. Structural estimation reveals that this policy change altered the underlying data-generation process, and the cor responding changes in forecasting behavior indeed align with our model’s predictions. Finally, we characterize how this rational confusion mecha nisminteracts withbehavioral biases suchasoverconfidenceandshowthat this interaction can address the well-known persistent forecast error puzzle in the literature.